Genplot Enhancements Log File
LIST OF ENHANCEMENTS IN ANSI-C DEVELOPMENT
<News>
<item> Windows users, the new MESSAGEBOX command may be very useful.
See MessageBox -? for further information and help.
<item> Major changes to the string handling code was implemented in Fall
   2007.
</News>
===========================================================================
Prty    0:      Nice upgrade feature
        1:      Minor bug but not severe, or high priority feature upgrade
        2:      Major annoying bug but not computationally wrong
        3:      Error in results -- must be fixed now
===========================================================================
MOT 8/7/2013 - Added ability to change color of axes (just one)
     AXCTRL <axis> -COLOR <n>
  Will cause the specified axis to be drawn in another color.  The settings
  of $AXCOLR[] will override these settings, so can force the title to be
  in black if desired (or tick marks)
     axctrl left -color 2
     let $axcolr[4] = 1
     axis
  Note that all four axes are independent with respect to color.  Although
  the right normally follows the left characteristics (unless on), this will
  not happen with color.

MOT 8/1/2013 - New smoothing capability
  It isn't often that I'm surprised by GENPLOT anymore.  But smoothing of
  square waves turned out to look really ugly.  Learned that the FFT_SMOOTH
  is really a convolution with an inverted bell rather than with a true
  Gaussian.  So had to implement the Gaussian smooth to deal with curves
  containing very high spatial frequency "signals".  Expanded SMOOTH with
  options to handle, as well as new commands
     SMOOTH -FFT <pts>      - Smooth by the inverted Bell over <pts> width
     SMOOTH -GAussian <pts> - Smooth by true Gaussian over <pts> width
     GAUSS_SMOOTH <pts>     - Smooth by true Gaussian over <pts> width
     SMOOTH_GAUSSIAN <pts>  - Smooth by true Gaussian over <pts> width
     FFT_SMOOTH <pts>       - Existing function using inverted bell
     SMOOTH_FFT <pts>       - Existing function using inverted bell
  The effect of <pts> is the same in both cases, reducing Gaussian noise
  by a factor of approximate 1/sqrt(pts).

MOT 7/10/2013 - Statistics on probability distribution functions
   The new functions
       @pdf_ave(cv [,xl,xh])       @pdf_avg(cv [,xl,xh])
       @pdf_average(cv [,xl,xh])   @pdf_mean(cv [,xl,xh])
       @pdf_median(cv [,xl,xh])
       @pdf_rms(cv [,xl,xh])
       @pdf_std(cv [,xl,xh])       @pdf_sdev(cv [,xl,xh])
       @pdf_stdev(cv [,xl,xh])     @pdf_stddev(cv [,xl,xh])
       @pdf_variance(cv [,xl,xh])  @pdf_var(cv [,xl,xh])
       @pdf_skew(cv [,xl,xh])      @pdf_skewness(cv [,xl,xh])
       @pdf_kurt(cv [,xl,xh])      @pdf_kurtosis(cv [,xl,xh])
   treat a curve as a probability distribution histogram.  The moments
   of this pdf (as an integral) are computed.  If the optional range
   is specified, the curve is assumed to be sorted in X.

MOT 7/10/2013 - Documentation of behavior
   The statistical commands @mean, @skew, etc. allow a curve as well as
   an array as the argument.  If a curve is given, the Y argument is
   treated as the array to scan.  The optional range arguments, which
   are integer indices into the array, become X values for the array.
      eval @mean($plot, -9.5, 9.5)
   returns the average value of the default Y array but only including
   points where -9.5<x<9.5. The range is optional and defaults to the
   entire domain.

MOT 6/16/2013 - Added ability to "uncurl" an AFM image
    The AFM images often come with a complex "curl" in X, Y or both,
    where curl is a "techical" term for a smoothly varying background
    that is not removed by the flatten command.  In either X or Y, the
    average height is fit to a polynomial and then subtracted from the
    data.  Some day, "uncrumple" may be added to remove interactions
    between X and Y, but currently only uncurl.
       matrix s1 uncurl x 6
    uncurls the surface s1 along the X direction using a 6th order
    polynomial fit.

MOT 4/11/2013 - Added conditions to properly read a new format of DI
    AFM format files.  From the CNF cleanroom.

MOT 3/14/2013 - Added more functions - Fermi-Dirac integrals of useful
    orders.
      fdm0p5(x) - F_{-0.5}(x)
      fdp0p5(x) - F_{0.5}(x)
      fdp1p5(x) - F_{1.5}(x)
      fdp2p5(x) - F_{2.5}(x)
    From ACM bulletin - accurate to at least 6 decimals

MOT 2/25/2013 - Modified most of the @max type functions to allow
    operation on either arrays or curves.  If the argument is an
    array, the two optional arguments are interpreted as an index
    range to limit the data (inclusive).  If the argument is a
    2D curve, the two optional arguments are interpreted as X
    values to constrain the data.
        @max(c1,-0.5,0.5)
    is now equivalent to
        retrieve c1 
        cull keep xrange -0.5 0.5
        eval @max(y)
    Makes life much nicer when scanning for baseline levels in
    complex spectra.

MOT 1/4/2013 - Large file support.  Increasingly dealing with files
    more than 2 GB in size.  This is the 2^31 limit in non-64 bit
    architectures.  For interim, have fixed the "ll" command and the
    "filesize()" functions to properly handle large files (64 bit
    limit).  This should also fix some of the date problems that will
    start cropping up about 2037 (just about the time I get to start
    thinking about retiring).

MOT 1/3/2013 - Fixed the ^C handling in much of the code.  In both 
    XGenplot and Genplot, ^C now terminates command line, file, and
    block level repeat operations.  Attempts to flush input buffer
    so that the prompt for terminating running blocks is clean.

    Modified several commands/routines to limit potential delays:
        sleep / read / plot / let (long array evaluations)

MOT 1/2/2013 - Added visibility to the SysDebugFlag internal variable
    as the function variable $SysDebugFlag.  This is a bit-wise parameter
    that can be coded to provide detailed debug information during run.
    Generally not useful to common user ... but good for programmer.
        0x02 -> In gvcalc(), prints the pseudo-op codes from calculations
        0x04 -> In gvcalc(), also prints the stack and arg values

MOT 1/1/2013 - Fixed problem with wait timeout.

MOT 12/26/2012 - Added all the ctype.h functions for determining the type
    of character.  
          isalnum(c), isalpha(c), iscntrl(c), isdigit(c), isgraph(c), 
          islower(c), isprint(c), ispunct(c), isspace(c), isupper(c),
          isxdigit(c)
    and the two to convert single characters upper and lower case
          tolower(c)
          toupper(c)
    where c is any integer value (converted to nearest integer).
    Return value is system dependent, but non-zero if value is 
    of the specified class.  Behavior of these functions with values 
    of c outside the range 1 < c < 127 is also system dependent.

MOT 12/4/2012 - Added "DARROW" to draw a double arrow in annotage
     (you're welcome)

MOT 12/4/2012 - Added "align" as an alias for "orgmode" in Annotate.

MOT 11/26/2012 - Added options -DEBUG and -DETAIL (synonymous) to the
      LISTVAR function.  Also added -? to give help.  The debug mode
      prints out internal information to help debug operation.

MOT 11/15/2012 - Added rnd_norm() as a simple way to get a normal
      distribution.  Changed rnd_normal() so that it requires giving
      two parameters (mean and stdev)

      Ian suggests that GVPARSE should be revised so that there is a way
      to allow either zero or all of the parameters in a function defn.

MOT 11/4/2012 - Added @E(x) and @cov(x,y) as function in analog to @var.
      @E is expectation value (mean), @cov is covariance

MOT 10/31/2012 - Added -HBARGRAPH to the plot options.  Horizontal bar
   graph.

MOT 10/28/2012 - Added normal() as alias for gauss, and new functions for
   chi squared distributions.
        normal(x,mu,sigma)
        chisqr(x,nu)                /* Complex numbers use only real part
        chi2(x,nu)                  /* Complex numbers use only real part
   This function has the expected means and distributions

MOT 10/24/2012 - Added the @MAD(array) and MAD(x1,x2,...,xn) functions to 
   return the MAD of a sample.  The MAD is the "Median Absolute Deviation" 
   and is an analog to the standard deviation using median concepts.  Also
   fixed median() function to properly average when the number of items was
   even.

   For complex numbers, the sorting for the median is based on the real
   value first, and if the real is equal, on the imaginary part.  However,
   for the MAD, the magnitude of the argument |z-median()| is used.  It
   really doesn't make sense to do median/MAD of complex numbers, but do
   so nonetheless.

MOT 10/17/2012 - Added a "NHISTOGRAM" and "CDF" transformations to the
   transformation function.  The NHistogram is a normalized histogram
   which automatically sets the -densify and -normalize options.  This
   makes it easier to work with probability function estimation from data.

   The CDF transformation generates the Cummulative Distribution Function 
   which is the integral of the probability distribution.  F(a) = P(x<=a).
   Again, a function useful for statistical analysis.

MOT 10/6/2012 - Added a "kernal" estimation for the probability density
   function capability.
      transform [y | x] kernal [-options] <bandwidth>
   This is a "idealized" histogram function that more smoothly estimates
   the probability distribution function for the samples in Y.  It gives
   a smooth curve of at least 500 points which mimics the histogram
     transf y histogram -density -normalize 
   but with continuous curves.  Default is use of the triangular kernal
   function.  Formally, the kernal method spreads each measured point out
   a the kernal function (think triangle) and then just sums up the values.
   This gives big peaks at values with lots of points and small peaks 
   where there were no results.

MOT 10/6/2012 - Changed histogram defaults.  In past, the default was to
   generate 100 bins for a histogram.  Have modified that to use a value
   that scales more reasonably with the number of points to bin.  Value now
   is 
           1.2 * @sdev(y) / npt^(1/3)
   From stats texts, the recommended constant is 3.49.  However, I find this
   creates far too few bins to see useful trends, so my default is 1.2.

   A fixed number of bins can still be specified using the -NX option.

MOT 9/18/2012 - Added Mersenne Twister algorithm for random number
   generation.  This is a the "best" current PRNG availalbe.  Implemented
   also random numbers from common distributions using the Mersenne Twister
     random()                   - Mersenne Twister random on [0,1].  Same as rnd_drand()
     rnd_seed(iseed)            - Initialize Mersenne Twister with iseed (default=time())
     rnd_lrand()                - Random integer value (32 bits)
     rnd_drand()                - Random double value on [0,1])
     rnd_iuniform(min, max)     - Uniform integers on interval [min,max]
     rnd_uniform(min, max)      - Uniform doubles on interval [min,max]
     rnd_exponential(mean)      - Exponentially distributed random numbers with specified mean
     rnd_erlang(k, mean)        - k-Erlang distributed random numbers with int p, double mean
     rnd_weibull(shape, scale)  - Weibull distributed random numbers with shape/scale parameters
     rnd_normal(mean, sigma)    - Normally distribution random numbers with mean/sigma
     rnd_lognormal(mean, sigma) - Lognormal distributed random numbers where ln(x) has normal mean/sigma
     rnd_triangle(min,max,mode) - Triangular distributed random numbers on [min,max] with mode
  The function rnd() is now mapped to rnd_drand() as the best implementaiton.

MOT 2/6/2012 - Added ability to set surface elements.  The form is
      let s1[x,y] = value
   where x and y are the coordinates of the point.  Will set the value into
   the closest element of the array given the specified x and y points. 
   This is guarenteed to be safe for all values of x and y ... being limited
   to the actual matrix.  x and y are compared against s1:x and s1:y and need
   not be integers.  If s1:x or s1:y are not strictly sorted (either ascending
   or descending), the behavior is undefined but safe.

   Because of the , in the format s1[x,y], the use of "let" is required to
   set the element value.  Implicit let is not available currently.

   This command now also will force me to create a new variable class of 
   matrices, including potentially n-dimensional as well as 2-dimensional.
   This function allows working with matrices, but the (x,y) format is
   backwards to the (row,column) typically expected for matrix manipulation.

MOT 2/6/2012 - WARNING (mostly to self): The change above required an order
   switch in math parsing of expressions to allow a )]} character to be a
   delimiter in a parsing string.  Previously, the check for open or close
   parenthesis was done before checking the delimiter list, now the delimiters
   are checked first.  Should have no effect except to enable the s1[-7,3] to
   be properly parsed, but always nervous.

MOT 2/3/2012 - Extended function evaluator to be able to determine values
     from a surface.  Given s1 as a surface, s1(x,y) returns the value of
     the surface at that point, interpolating as necessary.  If the surface
     x and y values are left as indices, this can be used to address the 
     surface as a matrix.
          alloc s1 surface 20 30
          eval s1[0,0]
          eval s1[18,10]
    This is equivalent to the @zinterp(s1,x,y) function which is now
    obsolete (deprecated).  From the @zinterp original implementation:

    Attempts to interpolate between points on a surface.  Uses constant
    extrapolation beyond the edges of the surface.  Attempts to properly
    handle X,Y scales on surfaces, but don't push too hard (ie. unsorted).
    Basically gives an analytical way to determine values off of a
    surface for complex functions.

MOT 1/14/2012 - Added help to USER and LOAD commands.  Forgot how to use
     the LOAD command.

MOT 11/11/2011 - Removed "fit power" as an alias for "fit poly".  Will ultimtely
       add a new fit capability for true power law relationship (y = ax^m).

MOT 11/11/2011 - Added help to FIT and some of the FIT commands (for students at
       Cornell)

MOT 11/11/2011 - Fixed problem in WordPos() function.  Addition of ability
       to scan for phrases as well as single words had unintended consequences.
       Now fixed.

MOT 11/11/2011 - Version update to 2.11.  (Wanted 11/11/11 so a few days early)
       Added two new variables to be able to track the version
           $version - real number representation of the version
           $version:major - integer major version (2)
           $version:minor - integer minor version (11)
       Allows for version testing in the case of incompatibilities.

MOT 11/3/2011 - Added "fit constant [-sigma ... -range ...]" function to
     the fitting list.  Really implements the same as @wave(x,s) but in a
     way that is perhaps more obvious.  Same options as fit linear.

MOT 8/15/2011 - Added high precision timer() function.  This function under
     Windows has sub-microsecond resolution for timing events.  Under 
     other operating systems, currently is equivalent to time() function,
     with optional reset.
            timer([reset]) - optional BOOL reset parameter
            timer(1) ==> reset timer
            timer()  ==> get time since last reset
            timer(0) ==> get time since last reset

MOT 7/15/2011 - Modified "alloc <var> string <length>".  If the length is
     specified as non-zero, then a string of that length will be created 
     and filled with blanks (plus one char for the terminating \0 null).  
     If the length is speciifed as zero (or negative), the variable name
     is defined but the string is NULL (not empty, but invalid).  Any
     subsequent "let <var> = " will properly define the variable.  In all
     cases, the string will always expand to the required size on a let
     command.  In previous versions, all values of the length were ignored
     and the string was not initialzied on an "alloc" command.  The only
     impact is that the string may be used immediately in expressions if
     allocated as a true blank string, while it will give an error if left
     as a NULL pointer.  The default value for an "alloc" command is 1.

MOT 7/15/2011 - Tired of normalizing the Gaussian when I want to know the
     peak.  So created new function
         gaussn(x,x0,dx)
     for normalized gaussian to a peak of 1.0.  Just removes the 1/sqrt(2*pi)sigma
     normalization factor in front of the Gaussian distribution.

MOT 7/15/2011 - Tired of having to count lines in data files with text marking
     the data.  Added options -begin <text> and -end <text> to Genplot read
     command to allow read to being past first line containing the text, and
     terminate on the first line containing the end text.
         read 110712_171334.dat -begin "[BEGIN_DATA]" -end "[END_DATA]" -col 0 1 
     The comparison ignores whitespace at the start of the line, and is
     case insensitive.  Must be able to block data with specific comments such
     as above to use.  Otherwise, go back to counting and use the -rows or
     -lines option.  

     This may be used with the -rows option.  Row numbering starts at the first
     line past the -begin <text> line ... essentially lines prior to (and
     including) the identified text line are ignored.  Reading terminates on 
     the first line containing the end text.

     Default is to not ignore any text.  And obviously this has no effect on
     any non-ASCII read.

MOT 7/15/2011 - Added -lines synonym for -rows in GENPLOT read command

MOT 7/15/2011 - Everything else in GENPLOT is 0 index based.  Change reading
     AVI files so the same.  Frame 0 is now the first one.  So as not to break
     old macros (but still may), code allows the last frame to be referred to
     as $FrameCount-1 (correct) or $FrameCount (incorrect but previously used).

MOT 7/7/2011 - The functions magn(), imag(), real(), conj() and arg() now force
     the function evaluator into complex calculation mode.

MOT 6/21/2011 - Modifications to REXX.  During documentation of the
     functions, a number of *corrections* and enhancements have been made.

     ERROR in implementation fixed:
        lastpos(needle, haystack, start) - start was handled wrong.  It
              appears that it should be the starting position within the
              string of the backward search.  It will be 1 based, with 1
              being essentially strcmp() (only search starting at the first
              character) and strlen(haystack) being the default (start 
              search at the last character in the string).  However,
              returns a 0 based position in the string.
     Added/changed - all functions of REXX that make sense are fully implemented.
                     all share issue of 0-based counting in GENPLOT versus 1 in other REXX
                     functions insert and xrange have specific deviations from standard behavior
     REXX string functions
      abbrev(s1,s2,n)                      center(str,n [,c1]])                 centre(str,n [,c1])
      compare(s1,s2 [,pad])                copies(str,n)
      delstr(str,n[,len])                  delword(str,n[,len])                 index(s1,s2 [,n])
      insert(str,new[,posn[,len[,pad]]])   justify(str,len[,pad])               lastpos(s1,s2 [,n]
      left(str,n [,c1]])                   length(str)                          overlay(str,new[,posn[,len[,pad]]])
      pos(s1,s2 [,n])                      reverse(str)                         right(str,n [,c1]])
      space(str[,n[,pad]])                 strip(str [,mode [,chr]])            substr(str,n[,len[,pad]])
      subword(str,n [,len])                translate(str[,new,old[,pad]])       word(str,n)
      wordindex(str, n)                    wordlength(str, n)                   wordpos(s1,s2 [,n])
      words(str)                           verify(s1,ref [,mode [,start]])      xrange(istart, iend)

     These functions are also now specifically listed in the "eval -list" command.  eval -detail REXX
     will give detailed usage.

     substr() is widely used and its behavior now conforms to REXX.  In particular, the third parameter
     for the length of the substring should normally be unused.  If given, the resulting string will
     be exactly that size long, padded with (default) spaces.

MOT 6/21/2011 - Added code to handle special cases of stdin, stdout and
     stderr in the file handling routines.  These can now be used instead
     opening a specific file pointer.
         fgets(stdin)
     will read from the keyboard (no echo) returning on the <nl>.  Likewise
         fputs("line 1: ...", stdout)
     will print the string to the console with no additional formatting.

MOT 6/14/2011 - Added the function printf() synonymous with fprintf() but
     outputs directly to the console.  Exact same format and same required
     use of control characters for newline.
            printf("format", var1, var2, expr1, expr2)

MOT 6/14/2011 - CHANGE IN BEHAVIOR OF FPUTC and FPUTS
     The order of arguments for these functions is now consistent with the
     C language equivalents.  The stream pointer comes last rather than first.
        fputc(ichar, funit)
        fputs(text, funit)
     Since adding more flexibility in argument handling, this was easy to do
     and makes more sense than being inconsistent with the existing language.
     Apologies to the few that it will impact.

MOT 6/14/2011 - CHANGE IN THE BETAI FUNCTION.  CRITICAL.
     The function betai(a,b,x) did not implement the incomplete beta
     function B(x;a,b), but rather the regularized incomplete beta
     function I_x(a,b).  Have changed the order of the variables and
     added a new function to implement both
        betai(x,a,b)    ==> B(x;a,b)  Incomplete beta function
        betai_Ix(x,a,b) ==> I_x(a,b)  Regularized incomplete beta function
     It is the regularized incomplete beta function used most extensively
     in statistics.

MOT 5/28/2011 - Modified the solve() function to not fail immediately if
     the signs of the function at the specified boundary edges are the same.
     Instead, try up to maxiter random positions within the range looking
     for a sign reversal.  Since a random search is made in the interval,
     there is no guarentee that the function will return the same root each
     time, but it is guarenteed to return a root (or an error).  Using a
     random search reduces the chance of missing a zero with periodic 
     functions.

MOT 5/28/2011 - Worked a bit on the logarithmic labeling.  Now draws 
     tertiary labels for short logarithmic axes.  Down to the level of
     1.1, 1.2, ... for very small ranges (which is really stupid).  Also
     redid autoscaling so it works much more responsibly.

MOT 2/27/2011 - Increased the maximum length of a continued line to 16384 
     characters.  Also changed alias code so it could accomodate a
     definition with almost this length.

MOT 2/15/2011 - Added extensive list of special characters draw capability.
     Basically added the entire symbol font code page where I could identify
     the symbols and give them LaTeX like definitions.

MOT 1/10/2011 - CHANGE TO @f_test(ar1,ar2) function *** IMPORTANT ***
     I have concluded that there was a mistake in coding the @f_test function.
     It is now coded such that it explicitly returns the same as
          @f_test(ar1,ar2) = f_test(@var(ar1)/@var(ar2),ar1:npt-1, ar2:npt-1)
     This is reflected in the new (upcoming) PDF documentation.

MOT 1/12/2001 - Added digamma(x) function
     digamma(x) which is also known as psi(x) is the derivative of lngamma.
     Now implemented for real arguments only with precision ~10^{-15}
        digamma(x)
     Like lngamma, it is undefined for 0 and negative integers, and the function
     return TMPREAL_MAX.

MOT 1/12/2001 - Modified lngamma(x) function
    Now returns TMPREAL_MAX instead of 0 for 0 and negative integer arguments.

MOT 1/12/2011 - Added aliases for inverse trig functions
     asin  -> arcsin (same for acos,atan,etc.)
     asind -> arcsind (same for acosd, atand, etc.)
     asinh -> arsinh  (same for acosh, atanh, etc.)  [Note only arsinh, no arcsinh]
    These made sense to do as writing the new documentation.

MOT 1/5/2011 - Added ability to "beep" from the code
       _beep(freq, durat)
     will beep.  Both parameters are optional and default to 880 Hz (A)
     and 150 ms.  For fun, try
        foreach (0,2,4,5,7,9,11,12,11,9,7,5,4,2,0) qev beep(220*2^{(3+%f)/12},200)

MOT 1/4/2011 - Low level file I/O for working with serial ports (and most
     USB instruments).  
	_open_comx(int port [,string baud [,int ms_timeout]])
        _get_baud()
        _set_baud(string baud)
        _get_timeout()
        _set_timeout(RdInterval, RdPerByte, RdOverhead, WrByte, WrOverhead)
     _open_comx() opens the specified COMx port as a serial Low I/O stream.
     The optional baud parameter is a text string describing the flow
     parameters, such as "9600,n,8,1".  It is parsed by the OS to set 
     parameters.  The third (optional) parameter is a timeout setting for
     the _read() command in ms.  The total time waiting is the number of 
     bytes requested plus this timeout (in ms).
       setv fd = _open_comx(10,"921600,n,8,1",2000) 
     opens COM10 with typical USB parameters and a 2 second timeout on read
     with no data.  Note that
       let aline = _read(fd, 2000)
     will timeout in 4 seconds (2 for the potential 2000 characters plus
     the 2 seconds specified).  Once a read receives a character, it will
     always timeout after 50 ms if no additional characters are received. 
     These parameters work very well for USB serial devices, but the code
     may need to be modified to handle slow devices (300 baud).  If the
     rate and timeout values are not given, they are not set and retain
     default values from the operating system.

     The string for setting baud/flow parameters in _open_comx() and _setbaud()
     is identical in format to that required by the Windows "mode" command.
     You should not set the com port, but only remaining parameters.
         [baud=b][parity=p][data=d][stop=s][to={on|off}][xon={on|off}]
         [odsr={on|off}][octs={on|off}][dtr={on|off|hs}][rts={on|off|hs|tg}]
         [idsr={on|off}] 
     The older format works as well.  "9600,n,8,1" to specify a baud rate of
     9600 with n parity, 8 data bits and 1 stop bit.

     The _get_baud() and _get_timeout() functions print (do not return) the
     current values of all the communication parameters and the members of
     the timeout structures.  These can be individually modified using the
     _set_baud() and _set_timeout() functions.  For _set_timeout(), the
     default for any parameter is for no change.
       _set_timeout(,,1000)  <== only changes the read overhead time

     Consider these calls to be fragile and that they may change.
       setv fd = _open_comx(10,"921600,n,8,1",500)
       qev _write(fd, "1PA?")
       eval _read(fd,20)
       qev _close(fd)

MOT 1/4/2011 - Low level file I/O capability including serial ports (Win only)
     Added low level I/O functions _read(), _open(), etc. and associated
     constants.  
	_open(char fname, int flags [, int mode])
	   O_CREAT  O_APPEND O_TRUNC
           O_BINARY O_TEXT
	   O_RDONLY O_RDWR   O_WRONLY
	_creat(char fname [,mode])
	_close(int fd)
	_read(int fd [, int maxcount])               maxcount defaults to 256
	_write(int fd, str string)
	_query(int fd, str string [, int maxcount])  maxcount defaults to 256
	_lseek(int fd, int offset, int whence)
	   SEEK_SET SEEK_CUR SEEK_END
	_eof(int fd)
	_tell(int fd)
     In general, these directly convert to the equivalent C routines except
     for _read() and _query() which return a string up to the size specified.
     _query is a sequence of _write() and and and immediate _read().

MOT 10/12/2010 - Added ability to generate arbitrary (text based) labels to
     the graph.  This replaces the automatic labeling.  Must specify lots of
     values to use.
          Values of the major/minor tick marks   (REAL arrays)
          Labels at the major tick marks         (STRING arrays)
          Label size (0 ==> use graph max)       (REAL)
     This permits more complex non-linear axes (Weibull probability) as well
     as non-numeric labels (calendar months Jan, Feb, ...)
         alloc s1 string_array 12
         foreach ("Jan" "Feb" "Mar" "Apr" "May" "Jun" "Jul" "Aug" "Sep" "Oct" "Nov" "Dec") let s1[%i] = %f
         alloc m1 array 12 let m1 = i+1
         alloc m2 array 100 let m2 = (1+i/4)
         axctrl bot -user m1 m2 s1 0.0
         label bot "Month"
         reg bot 1 12 force bot yes 
         axis
     or for labeling a non-linear axis
         xtop nonlinear
         alloc m1 array 12
         alloc s1 string_array 12
         alloc m2 array 100
         foreach (-20 -10 0 10 20 30 40 60 80 100 150 200) let m1[%i] = %f let s1[%i] = "%f"
         let m2 = 0 
         foreach (-25 -15 -5 5 15 25 35 45 50 55 60 65 70 75 80 85 90 95 100 110 120 130 140 160 170 180 190 210 220) let m2[%i] = %f

         xtop nonlinear
         label bot "1000/T (K^{-1})"
         label top "Temperature (\deg C)"
         label left "Rate (m/s)"
         axctrl bot -nouser
         axctrl top -user m1 m2 s1 0.0

         reg bot 2 4 force bot yes
         reg left -1.0 0.0 
         alloc m3 array 10
         alloc m4 array 100 let m4 = log((i+1)/100)
         alloc s2 string_array 10
         foreach (0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0) let m3[%i] = log(%f) let s2[%i] = "%f"
         axctrl left -user m3 m4 s2 0.0
         axis

   
MOT 10/2010 - Added @covar and @covariance to statistical tests.
       @covar(ar1, ar2 [, ul, uh])
       @covariance(ar1, ar2 [, ul, uh])
     The covariance estimate is 1/N-1 sum (x-\bar x)*(y-ybar y) and is a 
     measure of correlation in error between two measurements.

MOT 7/2010 - Modified LexGetFile to allow , and ; in filenames without use
     of quotes.  Should have been done long ago.

MOT 7/2010 - Added code to catch attempts to set NPT beyond NPTMAX on 2D and
     3D curves.  Works for curve:npt and genplot's NPT when set by the command
     LET (either implicitly or explicitly).  No more crashes when making NPT
     bigger before increasing the curve size :-)

     The curve will be automatically resized to 10% beyond the value if
     possible (since resize is marginally expensive operation).  Gives error
     and sets NPT to NPTMAX if the resize cannot be done.

     GENPLOT: npt = 10000
     INFO: NPT (10000) requested is larger than NPTMAX.  Increased curve size to 11000
     GENPLOT: npt = 100000000
     ERROR: Attempt to set NPT (100000000) to more than GVI_MAX_LENGTH (67108864).
            Value set to nptmax (11000)
     GENPLOT:

MOT 7/2010 - Added statistical functions to return the mean and sigma when the
  the uncertainties (or relative uncertainties) are known.  Referred to as the
  weighted mean and related values.
      @wmean(x, sigma [,ilow, ihigh])     Returns weighted sum, where weighting
      @wavg(x, sigma [,ilow, ihigh])      is 1/sigma^2 for each point
      @wave(x, sigma [,ilow, ihigh])
      @wabsavg(x, sigma [,ilow, ihigh])   Weighted sum of the absolute value of x

      @wsigma(x, sigma [,ilow, ihigh])    Uncertainty in the mean given known and
                                          valid uncertainties.  This is equivalent to 
                                          the standard deviation of the mean in a well
                                          defined statistical environment.

      @wvariance(x, sigma [,ilow, ihigh]) Variance as determined from the data itself
      @wvar(x, sigma [,ilow, ihigh])      using sigma only as the relative weights

      @wstd(x, sigma [,ilow, ihigh])      Standard deviation estimated from the data
      @wstdev(x, sigma [,ilow, ihigh])    using sigma only as the relative weights.
      @wstddev(x, sigma [,ilow, ihigh])   If sigma's are valid, should be average sigma

      @wsdom(x, sigma [,ilow, ihigh])     Estimate of the standard deviation of the mean
                                          using sigma only as the relative weights.
                                          If sigma's are valid, should almost equal 
                                          @wsigma(x, sigma)

MOT 6/2010 - Set maximum number of arguments in a function to 32.  This
             has always been the case but never checked.

MOT 6/2010 - Added ability to specify an array or any other object in the
    math namespace as a potential variable in a function defition.
    This is really a string replacement that acts to be able to specify an
    array in the function definition.  The prefix * indicates that a variable
    should be taken as being an array.
        define f(x,*ar) = ar[0]+x*ar[1]
        alloc cp array 2 let cp(..) = 7,20
        eval f(7,cp)

        define f(*c) = @max(c:y)-@min(c:y)
        create y = sin(x) -range -15 15 eval f($plot)

MOT 5/2010 - Increased the buffer for fgets() from 4096 to 32768 chars.

MOT 5/2010 - Added % as a valid terminus on a number for percentage.
   GENPLOT: (1+7%/360)^360-1
    := 0.072500883

MOT 5/2010 - Added functions
       mantissa(x) - returns signed mantissa of a large number
       exponent(x) - returns exponent for scientific notation
    Common use  fprintf("%.3fE%d", mantissa(x), exponent(x))
       
MOT 4/24/2010 - Moved lsqfit into GPTFIT so that it shares the cf$[] array
    and sets both the values into the function evaluator and the function
    fit(x).

MOT 4/16/2010 - Added code so integrals of complex functions along the real
    axis are possible.  Note, the contour of the integral is limited to the
    real axis so upper and lower limits must be real values

MOT 4/7/2010 - Added ability to link to external functions that are functions
    of both real and string variables.  New linkages
       GVLinkFncA()
       GVLinkStrFnc()
    Used primarily with user routines to extend capability.  Needed myself
    to create a user module that permits control of GPIB instruments.

MOT 4/1/2010 - modified the solve() function work with double precision
    variable rather than single.  Get more precise and avoids problems 
    of significant digits.
          solve(tan(x)-1.28E5*x,0.99999*pi/2,(1-1E-8)*pi/2)
    Used to fail because (1-1E-8)*pi/2 too close to pi/2 numerically.

MOT 1/30/2010 - Added function to do summation and continued product over
    a function.
         sum(fnc(var)|var, ilow, ihigh [,istep])
         prod(fnc(var)|var, ilow, ihigh [,istep])
    For the sum and product, the assumed variable is i, not x.  Use the |
    format to specify some other dummy variable.
         sum(i,0,10) = 55    sum(i,0,10,2) = 30
         prod(i,1,10) = 10!
         sum(cos(2*pi*n*x)|n,0,100)
    The function is fully implemented for both real and complex functions.

    Example:
       identify -place 5.65 6.0
       reg bot -6 6
       reg left -1 1
       define f(x,n) = sum[ m1n(i)*cos((2*i+1)*x)/(2*i+1), 0, n-1 ] 
       axis
       foreach (1 2 3 5 10 20 50 100 200 500 1000) {
          ov -f f(x,%f) -range -6 6 -points 500 -lt 1 -identify "%f terms"
       }

MOT 1/29/2009 - Added standard deviation of the mean to the functions
       @sdom(array[,istart,iend])
       
MOT 1/12/2009 - Fixed problem with matrix transpose and added more
    matrix manipulation.  Matrix transpose failed to relink the X and Y
    arrays which created problems in cases where the pointers where 
    pulled by name rather than structre.  Now properly relinked.
    Added
            matrix put_submatrix <src> <dest> <row> <col>
            matrix submatrix_put <src> <dest> <row> <col>
    to insert a matrix into an existing matrix.  Nice for making quilts.
    Also added aliases
            matrix get_submatrix
            matrix get_row
            matrix get_col
    
MOT 11/10/2009 - Added ability to write proper csv files.
       WRITE <fname> -csv
    will generate comma separated values (csv) rather than my normal
    tab separated values.  Allows the file to be opened in Excel without
    any further issues.

MOT 11/10/2009 - Added RDF capability for matrix evaluation
         MATRIX <surface> RDF [-center <x0> <y0>] [-bin <dx>]
    will generate the radial distribution function from the image in 
    <surface>.  By default, the origin of the curve is given by the zero
    in the X and Y values of the surface.  However, the option -CENTER will
    permit any value to be taken for center.  The RDF histogram by default 
    will be 1 average "pixel" of the image (good in most cases).  Any value
    of the bin width may be specified with the -BIN option.  y[i] of the
    result contains the average of all pixels in the image at distance
    i*dx <= r < (i+1)*dx.  Great for TEM diffraction image analysis.

MOT 6/28/2009 - Added ability to clip ANNOTE structures to the graph 
    area.  By default, annote is allowed to draw over the full canvas.
    Commands (such as line and arrow) that allow options now permit the
    option -clip to limit to the data area.
       line -lw 8 -clip 1 7 1 9 2

MOT 6/27/2009 - Added ability to specify an editor for the editor or eps
    commands.
         declare $editor = notepad
         declare $editor = wordpad
         declare $editor = "c:\program files\eps13\bin\epsilon.exe"
    will cause this to become the editor of choice.  Can be set in the
    genplot.ini file.

MOT 6/18/2009 - Added new palettes (thanks to Roger de Reus)
    More rainbow palettes
         GREY GOLD COPPER PINK AFM
         HOT  HEAT JET
         COLD COOL
         HSV BONE PRISM
         SPRING SUMMER AUTUMN WINTER
         MOT USER
    To see what these do:
       create -surface s1 z=x -range -1 1 -1 1 -rows 256 -col 256
       pl s1 -bitmap -palette hot

MOT 4/27/2009 - Change to matrix rotate function.  Prior code treated all
    surfaces as just a matrix and rotation assumed the same "step size" in
    X and Y.  Rotations failed to thus maintain proper dimensions when
    rotated with non-equal step sizes for rows and columns.  The code now
    properly handles different X/Y range, but still assumes uniform
    spacing of rows and columns across the X/Y range.  At some point, 
    may properly handle non-uniform spacing along X/Y as well, but not today.

MOT 4/21/2009 - Added function to return the last occurance of one string
    within another.  Optionally starts search offset characters from the
    end.  Note, if offset is less than or equal to the length of the 
    needle, it essentially has no meaning.
          lastpos(needle, haystack [,offset])
          lastpos("abra", "abracadabra") = 7
          lastpos("abra", "abracadabra", 5) = 0
 
MOT 4/21/2009 - Changed isdir, filesize and filedate (and isfile) 
    functions to more accurately deal with directory name conflicts.  A
    valid directory now is any string which appended with "/fname" forms
    a valid filename.  So in windows "C:" is valid, as is "C:////".  Any
    number of trailing slashes or backslashes are now acceptable (only
    slashses in UNIX).  Allows isdir("c:/windows/") to return true.  For
    files, these rules do not hold.  A file must be such that the same
    string could be successfully opened (as with a read).

MOT 3/25/2009 - Modified the graph window so window remains open when
    a new device is created.  Each time "DEV PM" is entered, the
    existing windows remains and a new window is opened.  The old one
    may be resized and manipulated -- but the data is static.  Windows
    that are inactive may be freely killed with the X button.  The active
    window will continue to complain.

    When Genplot or RUMP do exit, all windows are closed now.

MOT 3/25/2009 - Added code in devini() to allow the pm graph window to 
       (1) change titles based on how many calls have been made
       (2) exit cleanly once the main program exits
    These are related to the ability to have multiple dev pm windows
    open for archiving graphs.  Required change to the inidsp msgs.

MOT 3/25/2009 - Modified strip() to enable mode of removing all occurances
                of a character within a string.
                    strip(str [,mode [,char]])  mode = leading/trailing/both/all

MOT 3/25/2009 - Modified code so using "-ply right" works more as expected
       with a non-linear axes.  Will now automatically make the transform
       right_to_left(y) or top_to_bottom(x) to convert the data to the
       linear axes and plot as expected.  Effectively, the data is indeed
       plotted against the non-linear axis.

       This is guarenteed to work correctly with simple 2D graphs.  Haven't
       looked at how this might interact with drawing 3D graphs with
       the non-linear axes being selected.  If there are obvious bad
       hehaviors, will modify the code.

       Finally, there is no test that the transforms are valid.  Errors
       are probably set to zero values, but it is identical to the user
       doing the command "let y = right_to_left(y)".

MOT 3/25/2009 - Added functions for string oriented bitwise operations
       str = bitor(bstr1, bstr2)         str = hexor(hstr1, hstr2)
       str = bitand(bstr1, bstr2)        str = hexand(hstr1, hstr2)
       str = bitxor(bstr1, bstr2)        str = hexxor(hstr1, hstr2)
  Strings are zero extended to the left, and only the valid characters in 
  each string are considered.

MOT 3/25/2009 - Added functions that convert strings between hex and
       binary representations.
         hex2bin("str")
         bin2hex("str")
      The final string length is presently limited to 8192 characters.

MOT 3/25/2009 - Added more base conversion functions for Pat.  The full set
       is now:
         int2hex(int)       hex2int("hex_str")    hex_str may be 0x7F
         int2oct(int)       oct2int("octal_str")
         int2bin(int)       bin2int("binary_str")
         int2base(int,base) base2int("str", base)
       The conversion from int to a string is limited to values that
       are exactly representable as a simple integer.  -2^31 to 2^31-1.
       Conversion of a string to integer terminates on the first character
       of the string that is not valid in the specified base.
         bin2int("11031101") = 6
       For base 22 and above, recommend "upcase(int2base())" to avoid the
       potential confusion between "1" and "l" ("L").

MOT 1/6/2009 - Modified behavior of sorting for 3D curves to sort on
       both X and then Y as the default.  Added an option "-XY" which
       is explicit, but this is now the default behavior.  Sort follows
       X, and then if equal compares Y values.  It is not possible to
       do sorts on XZ or YZ presently.  Have to exchange the arrays
       first - sorry.  But -STRICT, -REVERSE, and -RANDOM are valid.

MOT 11/13/2008 - Added Z to the CULL command.  Now can cull based on Z 
       values in a 3D curve.  Only potential side effect is the RANGE
       mode for a 3D curve which now requests X,Y and Z whereas in the
       past would only request X and Y (even for a 3D curve).
       Also add a help to the command.

MOT 11/12/2008 - Expanded matrix command.
       MATRIX HISTOGRAM -- generates histogram from a surface.  Options
                           similar to TRANSFORM HISTOGRAM.  Use 
                           MATRIX HISTOGRAM -? to see usage
       MATRIX MARK -- more generic threshold generator.  Can window within
                      limited area, and change the values set in the Z
                      array.  Again, MATRIX MARK -? gives usage.

MOT 10/6/2008 - Added palette RANDOM to the list.  This palette divides the
     range into 1024 random colors, with exception that 0.0 is black and 1.0
     is pure white.  Useful for things like color maps of grain extraction
     from an SEM image.

MOT 8/11/2008 - Minor change.  Variable result$ is now a DOUBLE rather
     than a REAL.  Higher precision reflecting machines now moving to
     64 bit anyway.  Impacts only SOLVE and TRANSFORM INTEGRATE.  Should
     be completely transparent to users.

MOT 8/10/2008 - Using solve() to solve two simultaneous equations.
     Given requirement 
           f(x,y) = 0     g(x,y) = 0
     Idea is to use solve on f to determine x as function of y and then
     substitute into g and determine the y that satisfies the equation.
     As a specific example, consider
         f(x,y) = cos(x)-y = 0        g(x,y) = sin(x)-y = 0
     which has the obvious solution at x=pi/4, y = sqrt(0.5).  To be
     coded
         define xr(y) = solve(cos(x)-y,0,pi/2)    /* Gives x satisfying f(x,y)=0
         setv yroot = solve(sin(xr(y))-y|y,0,1)
         setv xroot = xr(yroot)
         eval xroot,yroot,pi/4,sqrt(0.5)

MOT 8/8/2008 - Warnings on dummy arguments with functions
     The dummy variable for functions of functions should normally not be
     used otherwise.  Many cases work, but other will fail as the program
     can't always figure out what is dummy and what is real.  Examples:
        create y = dydx(cos(x),x)         <== happens to works since the second x is an array
        create y = dydx(cos(xp)|xp,x)     <== safer and definitely more clear
        define f(x) = dydx(cos(x),x)      <== fails - what's an argument versus dummy
               eval f(pi/4)
	define f(x) = dydx(cos(xp)|xp,x)  <== works properly
               eval f(pi/4)

MOT 8/8/2008 - Implemented function to numerically integrate an arbitary function
        integrate(fnc,a,b,[,eps,mineval,maxeval])
           fnc      - arbitrary function, possibly with specified dummy variable
           a,b      - limits of the evaluation
           eps      - fractional change signifying convergence  (default = 1E-5)
           minevals - minimum number of evaluations of function (default = 256)
           maxevals - maximum number of evaluations of function (default = 65536)
     The algorithms is robust for almost any function, but not necessarily the most
     efficient for a specific function.  In the future, there may be an additional
     parameter specifying the algorithm.  

     Current algorithm is an extended Simpson's method based on recursive Trapezoidal
     rules.  Evaluation of the function is closed (end points are evaluated) on a
     uniformly spaced grid.  Because of the recursion, the number of evaluations will
     always be a power of 2.  The number of point is doubled at each iteration, until
     the fractional change in the estimate of the integral is less than eps.  The
     minimum number of evaluations is necessary to ensure that significant features of
     the function are not missed in a coarse grid (integrate(gauss(x,0,1),-100,100) and
     the maximum is to limit the total time.  For extremely complex but smooth
     functions, it is reasonable to reduce the minimum number of evaluations to as few
     as 16.

     The value of eps can be decreased for greater accuracy, but numerical stability
     has to be considered, especially for integrals that have an exact value of 0.0.
     Since a fractional accuracy is required, the number of iterations can increase
     to a point where the finite precision of the representation prevents convergence.
     For example,  integral(sin(x),0,2*pi,1E-6) requires 2^24 evaluations to converge
     and gives a less accurate answer than 2^16 evaluations.
            eval integral(sin(x),0,2*pi,1E-6,,2^16)
            eval integral(sin(x),0,2*pi,1E-6,,2^24)
	    eval integral(sin(x),0,2*pi,0.01)
     The first generates a convergence error, but is more accurate than the second
     which requires 16 million evaluations of sin(x).  The last, with only 0.01
     required relative precision gives 3.5x10^-16 with roughly 1024 evaluations -- the
     same result obtained 2^24 (16 million) evaluations.

     For most functions, specifying a precision below 1E-6 is unnecessary, but for 
     non-pathalogical cases doesn't invoke significant cost.  Convergence is O(1/N^4)
     where N is the number of evaluations so one additional "iteration" results in 
     a 16x reduction in the error.  For specific case integral(sin(x),0,m*pi):
         # evals     m=1          m=3          m=5          m=13          m=25
            4      2.0045598    2.8720866   -4.7868110   18.041038     50.113994
            8      2.0002692    2.0255403    2.2914919   -6.6257091   -18.404748
           16      2.0000166    2.0013951    2.0116391    2.1680170    -7.3241477
           32      2.0000010    2.0000845    2.0006641    2.0074576     3.1646882
           64      2.0000001    2.0000052    2.0000406    2.0004333     2.0305619
          128      2.0000000    2.0000003    2.0000025    2.0000266     2.0016485
          256      2.0000000    2.0000000    2.0000002    2.0000017     2.0000996
          512      2.0000000    2.0000000    2.0000000    2.0000001     2.0000062
         1024      2.0000000    2.0000000    2.0000000    2.0000000     2.0000004

     A very different issue arises in periodic functions where the data is undersampled
     and aliasing effects occur.  In these cases, the value may appear to be converging
     while undersampled, and then the value rapidly changes. Consider periodic function
     integral(sin(x)^2,-100,100)
            # evals     value
              4       17.725614
              8       17.760894
             16       17.76304
             32       17.763174
             64       17.763182
            128      127.41730
            256      100.45674
            512      100.43762
           1024      100.43671
           2048      100.43665
           4096      100.43665
           8192      100.43665
       16777216      100.43665
     After 64 evaluations, the value appears to be converging to 17.76 -- a direct
     result of undersampling the periodic function.  Only after 128 points is there
     sufficient "resolution" to see the full behavior of the function.  The worst
     possible scenario is represented by the function integral(sin(x),-30*pi,30*pi)

     Finally, some functions are poorly sampled with an equispaced grid -- such as
     1/x or similar functions with singularities.  A coordinate change can quickly
     improve the convergence and accuracy.
          int 1/x dx   let  u = ln(x)  yielding  x = e^u    and   dx = e^u du
          int_a^b dx = int_ln(a)^ln(b) du
     where the latter is obviously trivial.  A slightly less trivial example with 
     a similar result after the same substitution
          int [ 1/x sqrt(1+x^2) ] dx  =  int [ sqrt(1-e^2u) ] du
     and indeed
          integral(1/x*sqrt(1+x^2),0.0001,90)            = 98.897996
          integral(sqrt(1+exp(2*u))|u,ln(0.0001),ln(90)) = 98.897996 
     Hoever, the first requires 8 million evaluations to converge while the latter
     converges with only 128 evaluations.  Plotting the two functions makes it 
     obvious why the latter is so much more efficient.

     Bottom line: caveat emptor - let the user beware.  Know the basic behavior of
                  the function before using the integral function blindly.  It is
                  trivial to plot it over the interval first before doing the 
                  integral.  This may also suggest the possible coordinates change
                  required to improve the integration.

     Any of the optional arguments may be left out and the default will be used.
        integrate(cos(x)^2,-pi,pi,1E-10)       /* Higher precision
        integrate(cos(x)^2,-pi,pi,,32,128)     /* Fewer evaluations same precision
     Errors include inability to evalute the function or failure to converge.

     Using integral() in a create a curve is valid, but may not be the most efficient.
           create y = sin(x) -range 0 2*pi -points 201 transf y integrate
           create y = integral(sin(x),0,x) -range 0 2*pi -points 201
     are similar.  The first is much faster as it requires only 201 evaluations
     of the function, while the second typically requires 200,000 evaluations. 
     However, the first is increasing accurate for later points -- y[199] is a good
     approximation of the full integral while y[1] is probably pretty poor.  In
     constrast, the second is uniformly accurate (fractional) for all points.  The
     error at x=pi is 1.5E-4 (from the exact value of 1.0).

     Examples:   eval dydx(cos(x),pi/2)
                 create y = dydx(cos(theta)|theta,x)
                 eval dydx(f(x),0,1E-3)

MOT 8/8/2008 - Implemented function to numerically differentiate an arbitary function
        dydx(fnc,x[,dx])
           fnc - arbitrary function, possibly with specified dummy variable
           x   - point to evaluate the derivative
           dx  - optional argument specifying the "delta" for evaluation
     Simple numerical dydx(f) = (f(x+dx)-f(x-dx))/(2*dx).
     Examples:   eval dydx(cos(x),pi/2)
                 create y = dydx(cos(theta)|theta,x)
                 eval dydx(f(x),0,1E-3)

MOT 8/8/2008 - Implementing solve also allows implementing similar "functions
     of functions".  Have two more:
        dydx(fnc,x)                   derivative of function at x
        integral(fnc,xl,xh)           integral of a function between limits
        integrate(fnc,xl,xh)
     dydx is fine - simple numerical derivative.
     integrate is reasonable and may improve with time

MOT 8/8/2008 - Major new functions: solve(fnc,low,high) which will return the
     root of the function, assuming one exists between low and high.  This
     required considerable code modification and should be considered 
     "fragile" for the moment.  Allows full functionality of the SOLVE
     command as an inline evaluation.
             solve(fnc, xlow, xhigh)
             solve(fnc, xlow, xhigh, guess)
             solve(fnc, xlow, xhigh, guess, epsilon)
	     solve(fnc, xlow, xhigh, guess, epsilon, maxiter)
     Guess, epsilon and maxiter are optional but must be given in order.  If
     defaults are to be used, argument may be skipped.
             solve(fnc, xlow, xhigh,,,500)
     is valid.

     The fnc must involve a dummy argument, with "x" as default.  Any other
     dummy may be specified by following the function with |<arg>.  Examples:
         cos(x)             <== x is the dummy argument
         v/20+sin(theta)|v  <== v is the dummy argument
     The function can be arbitarily complex and may involve arrays or any
     other calls.  Don't get too fancy with string arguments however - remember
     these are still fragile.

     Examples:
         eval solve(cos(x)-x,0,1)                /* Simple use
         create y = sin(x) -range 0 1 -by 0.1    /* Array use
         eval solve(sin(xp)-y|xp,0,pi/2)         /* Should be 0,0.1,0.2,.. 

MOT 8/8/2008 - The parsing and evaluating of expressions was rewritten to
     be fully re-entrant.  Required to enable the solve(fnc,xl,xh) function.

MOT 8/8/2008 - Added operating system functions to function evaluator.
	chdir("dir")      |  cd("dir")
        mkdir("dir")
        rmdir("dir")
	rm("file")        |  unlink("file")
	mv("old", "new")  |  rename("old", "new")
    All return 0 if successful and -1 if unsuccessful.  No further error
    messages are available.

    NOTE: Unlink and rm aren't exactly the same, but for most will behave 
    the same.  At the program leve, use remove() or unlink() functions.
      
MOT 12/16/2007 - Added EVAL -?, EVAL -LIST and EVAL -DETAIL commands to be
       able to list all the internal functions available.  See eval -? for
       usage of the -LIST, -DETAIL and -APROPOS options.

MOT 12/11/2007 - Modified parse and eval functions so error during read will
       turn off after specifed $maxreadwarn.  Only valid for ASCII reads.
       Previously, parse or evaluation errors would continue to post even
       after the warnings from READ had been max'd out.

MOT 12/9/2007 - Removed the @t1_test(ar,mean).  Modified the @t_test function
       to accept two formats.
           @t_test(ar1, ar2) - Student t-test on two arrays of values
           @t_test(ar1, rval) - Student t-test against anticipated mean
       As these are the two most common uses of the t-test, it made sense
       to put in the code effort to detect the two usages.
       Others remain as defined.  Full list:
           @z_test(ar, mean, sigma) - Normal distribution comparions of
               array with known parent mean and standard deviation.  Returns
               the two sided probability
                      ndtr(|z|)-ndtr(-|z|) = 2*ndtr(|z|)-1
           @t_test(ar1,ar2)  -  Two sample t-test assuming the same variance
           @t_test(ar ,mean)  - One-sample t-test for specified mean.
           @u_test(ar1, ar2)  - Two sample t-test assuming unequal variance
           @td_test(ar1,ar2)  - Two-sample dependent T-test.  Used when
                repeated measurements are made of a sample population and
                existence of a difference is sought.  Tests (ar1-ar2) and
                compares mean with 0.

           All @t_test(a,b) functions returns the probability that a random set
           of samples from the same data set would have a t value less than or
           equal to the observed value.  99% ==> that only 1% of the time would 
           the agreement be this poor.  Read @t_test as the confidence that the
           NULL hypothesis (that the means are the same) can be rejected.

           Note: The @u_test does not truncate the DOF parameter to integer, but
                 uses the analytical extension of the Student t-test for real
                 values of DOF.  Thus values may differ slightly from other
                 implementations of this test.

MOT 12/7/2007 - Made leading whitespace ignored in datafile reads.  This
                applies to all input - leading and trailing whitespace is
                removed before any tests are done on the data lines.  Allows
                @end to be anywhere on line, as well as comments (/*, etc) to
                be indented without errors.

MOT 12/1/2007 - Increased maximum size of string in printf command to 8192 chars.
                This fills 100 lines on the screen with 80 characters / line.

MOT 11/29/2007 - Added functions to take list of values immediately
          ave(a,b,c,d)   average(a,b,c,d)   mean(a,b,c,d)
          std(a,b,c,d)   stdev(a,b,c,d)
          mean(a,b,c,d)
          count(a,b,c,d)
          @count(y [,ilow,ihigh])   <== just returns y:npt or ihigh-ilow+1 limited
    For complex arguments, average and stdev treat real and imaginary parts as
    independent.  For median, first sort on real and secondarily on imaginary.  
    Count obviously doesn't matter.  @count doesn't have a real/complex question.

MOT 11/21/2007 - Added option "-sum" for sort -strict modes.  Similar to -delete
    and -average, but totals up values in the equivalent X values

MOT 11/22/2007 - Added the @t1_test(ar,mean) function to complement the
               @t_test(ar1,ar2).  The @t1-test function tests an array
               compared to an expected mean value using the one-sample
               Student t-test.  The @t_test(ar1,ar2) uses the two-sample
               independent Student t-test assuming the same variance.  The
               @u_test(ar1,ar2) drops the assumption of the same variance.
                  @t_test(ar1,ar2)  - Normal 2 independent sample Student 
                      t-test on samples take from population assuming the
                      same variance.
                  @t1_test(ar1,mean) - One-sample Student t-test for having
                      specified mean.
		  @z_test(ar1, mean,sigma) - Z test on array for known mean
                      and standard deviation of the parent.  This is just
                      the normal distribution so returns two sided probability
                      ndtr(|z|)-ndtr(-|z|) = 2*ndtr(|z|)-1
                  @td_test(ar1,ar2)  - Two-sample dependent T-test.  Used
                     when repeated measurements are made of a sample
                     population and existence of a difference is sought.
                     Basically for arrays x,y looks at x[i]-y[i] and 
                     tests if the mean is consistent with 0.

MOT 9/9/2007 - Added config command "MACHINE_INFO" which just prints out
               the info returned by a uname() system call.  Useful in
               debugging, but not generally of user value.

MOT 9/7/2007 - Added ERRPRINTF command.  Similar to PRINTF command line
               but outputs with in the "error mode" with a beep and the
               alternate (default red) color.  Useful to say something
               went wrong.  For Windows users, there is also now the
               MessageBox command which puts up a system modal pop-up
               message box.  This can be used to really get attention.

MOT 9/7/2007 - Added MESSAGEBOX command.  Use MessageBox -? for help.
               This command is only available in the Windows version.

MOT 8/26/2007 - More matrix changes
                  pl -contour -at <zval>
               draws single contour at specified value.  Also increased
               the size of structures for a contour to permit up to 1E6
               points in 2000 chains.

MOT 8/26/2007 - More matrix changes
                  matrix flatten <surface>
               fits the surface to a plane and then subtracts off the
               non-zero part.  Leaves average of the surface unchanged
               while flattening out.  Does use the x,y values of the
               surface if given, but only impacts if non-uniform grid.

MOT 7/4/2007 - Curves and surfaces now define the internal variable :IDS
               pointing to the identifier string.  Consistent with all the other
               sub-element format.  The following commands are now synonymous.
                  let mycurve     = "an identifier"
                  let mycurve:ids = "an identifier"
                  retr mycurve let ids = "an identifier" archive mycurve

MOT 6/14/2007 - On PASTE_CLIPBOARD in XGENPLOT or XRUMP, any <TAB> characters
                are converted to spaces.  This will enable pasting of columns
                from Excel directly into a "read <<" command.  The fundamental
                origin of the problem is that a <TAB> is interpreted as a 
                request for filename completion on normal input.  When pasted,
                doesn't get handled properly.  Fortunately, cannot think of any
                reason that a paste operation would want that completion
                capability, or why a Paste would really need a <TAB> character.

MOT 6/8/2007 - Modified the handling of the "X" button on PMDRIVE.  It now
               gives a warning and message for how to restart the window
               if accidently closed.

MOT 6/8/2007 - Modified the handling of the "X" button on XRUMP and XGENPLOT.
               This button will now properly terminate the program with
               extreme prejudice, but only after warning the user.  You
               must select the RETRY to get it to fully close.  So far, it
               seems this will not leave the orphan processes in task
               manager so is fully effective.  However, nothing is done as
               far as cleanup or careful process termination.  Use with
               discretion.

MOT 6/8/2007 - Added ability to sort strings and arrays.  The "sort"
               command takes optionally a string_array or real_array
               argument and sorts with limited options.
                  SORT [curve | array | string_array] -options
                  SORT -curve   <name> -options
                  SORT -array   <name> -options
                  SORT -strings <name> -options
               Added the option -NOCASE which causes the sort to ignore
               case for strings.
                
MOT 6/7/2007 - Added reading of strings to the generic "read" of GENPLOT.
                 alloc c1 string array 500
                 alloc c2 string array 20
                 read test.dat -list c1 1 c2 2 /
               will work as expected.

MOT 6/4/2007 - Modified behavior of several array types (string, int, etc.)
               in both the allocated and linked version.  Should mostly have
               the :npt element now with the size.  The size can always be
               obtained with sizeof(), but now also :npt.  Using :npt allows
               the "size" to be actively changed, such as number of active
               elements in a string array.
               Have to be careful -- c1:x is an array, but c1:x:npt is
               not defined.  Only one level of "indirection" permited.

MOT 5/31/2007 - Added "GOLD" as a rainbow palette choice
                Added "USER" as a rainbow palette.  This uses an internal
                array $RAINBOW[] of 1024 elements as a lookup for the colors.
                   palette rainbow user                         /* Required to link $rainbow
                   setv nz = $rainbow:npt-1
                   define blue(x) = 255*(1-2*abs(x/nz-0.5))     /* New transformation
                   define green(x) = 255*(1-x/nz)
                   define red(x) = 255*(x/nz)
                   let $rainbow = rgb(red(i),green(i),blue(i))  /* Reset the $rainbow array from grey
                   create -surface s1 x^2-y^2 -range -1 1 -1 1 -grid 512 512
                   plot s1 -bitmap
                If you come up with a good color scheme, send me the eqns
                and I'll put it in as a hard-coded rainbow palette.  Suspect
                only power users will use this feature.

MOT 4/18/2007
   Added functions
      time2double(str)
      time2float(str)
   to convert a time string into seconds.  Handles only hh:mm:ss.s and
   does very limited error checking.  Terminates conversion on first
   invalid character.  Acceptable convertable numbers include
      18:24:16.427        - normal 24 hour time sequence
      7378:16.6           - minutes beyond 60 - acceptable with seconds
      918462.2            - just seconds with large number
      1827:24:17.7        - hours beyond 24 - acceptable
      26:84:93.6          - accepted, but pretty much meaningless

MOT 3/28/2007
   Added an option -nearest to 3d_grip to return the value of the
   nearest known X,Y,Z triple.  No interpolation.

MOT 3/1/2007
   Increased the maximum number of columns that can be read in a single
   READ -LIST <var> <col> <var> <col> from 40 to 128.  Still has a hard
   limit but hopefully now enough to get by most needs.

MOT 11/30/2006
   Implemented a -FOR <condition> into the possible limiting parameters
   of the general fitting functions.
      fit -linear -for abs(x)<5
   will now work.  Per request from Roger

MOT and IJT - 10/1/2006
   Added the commands PUSHD and POPD in correspondence with the command
   prompt in XP equivalents
     PUSHD - saves current directory and does as DCD (disk and drive change)
     POPD  - restores previously PUSHD'd directory
   PUSHD may be called multiple times, maintained as a link list.
   [Ian Thompson's first addition to the program -- next owner :-) ]

MOT - 7/18/2006
   By request increased string length for echo from 255 characters to
   LONG_STR_SIZE (currently 2048 characters)

MOT - 7/12/2006
   By request:
       LexEqual(str1, str2, minlen)
         minlen > 0  ===> string comparison is case insensitive
         minlen < 0  ===> string comparison is case sensitive
       absolute value of minlen is minimum number of characters that
       must match.
   Returns: 0 ==> not a match
            1 ==> matches all characters, but strings not identical
            2 ==> exact match (within case requirement)
   Leading spaces are always ignored in the strings.

MOT - 6/26/2006
   Added new matrix commands to find peaks in a matrix, and to put/extract
   fractional rows and columns
     matrix row_peak <source>
     matrix col_peak <source>
   Returns the maximum in each row or column (respectively) returning the
   data to the main curve.
     matrix extract_data <row> <col> [ROW | COLUMN]
     matrix insert_data  <row> <col> [ROW | COLUMN]
   These commands begin at the specified row/column in the matrix and either
   extract or insert the data from the main curve into the matrix.  Outside
   of establishing the start point validity, there is no checking on range
   validity of the insertion or extraction -- up to the user to preset the
   value of NPT and ensure that all references remain within the array.

MOT - 5/14/2006
   Removed most of the artificial limits on the 3D matrix read of
   unformatted data.  This applies to files that are just a sequence
   of numbers without headers indicating the size.
     read unformatted.3d.data -surface s1
   It will read up to a maximum of 33,554,432 (2^25) columns or rows, with
   a maximum of 2^26 data points (rows x colums).  This limit is set by the
   256 MB maximum for an allocated memory block in NT.  The read will
   actually continue until the memory allocation fails, so 64 bit versions
   of the OS may raise this limit.   The number of data objects on the first
   line sets the column dimension -- from that point on, format in the file
   is ignored and data is read sequentially assuming row by row ordering.

MOT - 5/12/2006
   Added some internal variables to control the GENPLOT read.
       $MaxReadWarn   - number of times a warning is printed on read
       $MaxReadErrors - number of errors allowed during read
          Warning messages are printed up toe $MaxReadWarn times,
          then just a message saying there may be more errors.  Read
          continues until $MaxReadErrors is reached, and then an
          error message is printed and read aborts.
          The previous behavior printed 20 WARNINGS and then just
          continued without messages forever.
       These variables are only linked on the first use of the READ
       command.  To be sure they are linked, a new option is added
          READ -DUMMY
       which does nothing but link the variables.
       [For Pat Smith and Roger de Reus]

MOT - 3/22/2006
   Added the option -PALETTE to plot/overlay which allows temporary
   setting of colormap basis for "rainbow" types graphs.  This includes
   the new bitmap drawing.

MOT - 3/22/2006
   Finally implemented the -BITMAP option in plot.  Works only for
   plotting a surface, but will draw it as a series of filled rectangles
   to the screen.  
         pl s1 -bitmap
   Ideally the axis should be drawn after a bitmap, but this breaks too
   many nice features in overlay.  Lets bitmaps be small regions of the
   screen as well.
   Useful options to -bitmap mode
      -zrange <low> <high>  - Overrides the range
      -grid <nrow> <ncol>   - Limits the number of elements in the bitmap.
            Every other / every third, etc. will be selected until then
            number to be drawn is less than specified.  Becomes an issue
            when a surface is on the order of 4000 points.
      -palette [afm | ...]  - see above, override the color palette

MOT - 3/21/2006
   The -rainbow option in plotting a 3D curve now uses continuum colors
   rather than the discrete 16 color previous.  Don't see a reason to
   maintain the old format.

MOT - 3/21/2006
   An issue of working with "subroutine" macros is the challenge of
   dealing with directories for the macros.  Modified XEQ command to
   enable "look for macro in directory of the current running macro".
      xeq c:\macros\test
        xeq <>/sub1   - Actually runs c:\macros\sub1
        xeq []/sub2   - Actually runs c:\macros\sub2
   Either <> or [] as leading characters is accepted.

MOT - 3/20/2006
   Got comment on painful printing of infinite error messages on some
   calculations.  I agree this is a major annoyance and have taken some
   efforts to eliminate.  Math evaluation will now print a maximum of
   25 error messages per "parse" of a function.  This won't eliminate
   all of the problems, but many.

MOT - 1/9/2006
   Many new functions to deal with 2D fft analysis of images and radial
   distribution functions.
      matrix peak_detect <surf>
         Returns curves with X,Y,[z] coordinates of peaks (higher than all
         8 neighboring cells).  Makes best estimate of the interpolated
         position within the space.  Fits row/col directions to quadratic
         and estimates peak position and height.  All 8 are used to
         establish that is a peak (must be absolute peak) but only 5 are
         used to estimate position (not fitting to all).  If two points
         are at exactly the same height (max), neither will be accepted
         as a peak.  Use "let z = z+0.001*ndtri(rnd())" to avoid this if
         it is an issue with quantized data.
      matrix threshold   <surf> <sz>
         Modifies surface so points >=sz are 1 and below are 0
      matrix window      <surf> <zlow> <zhigh>
         Modifies surface so points are 1 where zlow<z<zhigh, 0 otherwize
      matrix isolate     <surf> <sz> <cnt>
         Threshold at sz plus requirement of at least cnt points above
         threshold for along each column/row.  Isolates regions that
         would connected through only a one or two pixel "bridges".
      transform rdf <cutoff> <nbins>
         [-FAST | -MIRROR <3|2|1|0>]
         [-BOX <length>]
         Performs real space radial distribution operation on the
         2D or 3D data.  Each bin return contains density of point
         (per unit area) in an annulus (or shell) centered on the
         X value.  Total of <nbins> extending for 0<r<cutoff.  To
         handle edge effects, the data is assumed to have periodic
         boundary conditions, mirroring each point into the
         neighoring cells.  This dramatically slows down the
         generation of the radial distribution function - as much
         as a factor of 9 in 2D and 27 in 3D.  The -FAST option
         disables all mirroring and looks only at the points in the
         given data set.  This is okay fine as long as r<<L where
         L is the smallest dimension of the box.  The mirroring can
         be tailored with the -MIRROR <mval> option.
            <mval> |         2D           |          3D          |
           -------------------------------------------------------
              0    |   no mirroring       |   no mirroring       |
              1    |   mirror edges (4)   |   mirror faces (6)   |
              2    | + mirror corners (4) | + mirror edges (12)  |
              3    |   (same as 2)        | + mirror corners (6) |
           -------------------------------------------------------
         The default is 3, all mirroring.  This will give the proper
         rdf for r < L (minimum box size).
         The size of the box is automatically determined from the
         span of the data.  The -BOX option overrides this and
         enforces a PBC length (same for all dimensions).

MOT - 1/7/2006
   Incorporated reading of Digital Instruments AFM images into a
   surface directly.  This is not guarenteed to work in all cases, but
   it works reasonably well on the data I've dealt with.  Makes sense
   to put it generally available.
     read <fname> -AFM <image> [-frame <n>] [-rows ...] [-cols ...]
   The optional frame specifies the image within the file.  Default is
   to read the first image stored in the data file.  The rows and
   columns may be specified, but are presently ignored, and probably
   will stay that way.  Output is hopefully with all parameters in
   nanometers.

MOT - 1/7/2006
   Added matrix transformation to yield the S(Q) from the 2D FFT
      MATRIX S(Q) <surf>
   Takes the surface and generates the S(Q) averaged over for all
   equivalent Q vectors, assuming 0,0 is in the center of the image.
   This function is intended to be called only after a 2D FFT, though
   the math will work on any image.  Generates the spatial inverse of
   the radial distribution function S(Q)  (hence its name).
      2D
      MATRIX FFT s_data s_power -power
      MATRIX S(Q) s_power
      plot -lt 1
   Data near Q=0 is, as expected, noisy.  s_power:x values are assumed
   also to be the true Q values and are copied to the X coordinate of
   the resulting curve.  Curve size will be smaller of rows/2 or cols/2
   of the source matrix.  Again, expect rows = cols from the FFT.

MOT - 12/13/05
   Added the function
      @2DIntegral(<surf>)
   which evaluates the 2D integral of a surface structure.  Sorry, at the
   moment there is no subtlety to the integral - straightforward trapezoidal
   integration.  Each point is assigned an area equal to 1/2 distance to
   each x/y line.

MOT - 12/13/05
   Modified SETVAR to include ability to specify an integer type in sets.
    setvar [-int | -real | -double | -complex | -filetpr] <name> <value>
   Added -? as help also.

MOT - 12/1/05
   Added interpolation from surfaces.
      @zinterp(surface, x, y)
   Attempts to interpolate between points on a surface.  Uses constant
   extrapolation beyond the edges of the surface.  Attempts to properly
   handle X,Y scales on surfaces, but don't push too hard (ie. unsorted).
   Basically gives an analytical way to determine values off of a
   surface for complex functions.

MOT - 12/1/05
   Finished the 2D FFT capabilities.  Linked under the MATRIX commands
     matrix FFT <source> <dest> -options
        -power       Power spectrum                   (default)
        -magnitude   Amplitude/magnitude
        -dB          Power in dB
        -real        Real part of the FFT
        -imag        Imaginary part of the FFT
        -PSD         Normalized as a spectral density (per freq^2)

        -square      Square windowing function        (default)
        -parzen      Parzen windowing
        -welch       Welch windowing
   The windowing characteristics are important only if trying to look at
   dB.  Not nearly as useful as in 1D.  Must be better windows for 2D
   somewhere.

   Results of the FFT have the DC frequency at x[npt/2] (ie. x[512] in
   a 1024 point matrix.  Frequencies run from -fmax to +fmax-df.

   Matrix *must* be 2^n x 2^m in size.  Only tested extensively for
   square matrices - hopefully no issues if rectangular.  Spatial
   frequencies are correct based on the X/Y scales of the source data.

   Compare to DI AFM results, there are factors of 2 floating around.
   The maximum POWER and maximum AMPL are both 2X less than reported
   by DI.  This makes no sense (power should go as amp^2), so I choose
   to leave believing that I have it correct.

MOT - 11/30/05
   Added -BMP as alias to -BITMAP for read/write
   Added -COLORMAP as alias for -PALETTE in BITMAP write

MOT - 9/22/05
   Document: To set elements of an array
         let p(0..5) = 1,2,3,4,5
   You must use parenthesis () and not brackets [] and exactly two dots.
   But then you get to set multiple values in a single line.  Has been in
   the code for a long time, but never could remember the format.  Need to
   have a let -? options.  (SO DO IT)

MOT - 4/18/05
   Added the options -file and -name to read command.  These versions go
   through LexGetStrExpr instead of LexGetFile allowing use of string
   expressions for the name.
      read -name sprintf("die_%d.%3.3d", die_num, die_spot)
   will work properly.  The previous mode, not using -name, will fail
      read sprintf("die_%d.%3.3d", die_num, die_spot)
   with the message the "sprintf("die_%d.%3.3d", die_num, die_spot)"
   cannot be located as a file.

MOT - 4/8/05
   Modified the initialization of the print orientations so the clipboard
   and metafile default to rotated (normal viewing on screen and
   Powerpoint).  The values are not defaulted in the genplot_.ini
   distributed.

MOT - 3/15/05
   Modified the bitmap writing routine so that it corresponds to the X/Y
   values if they exist.  X will increase left to right, and Y bottom to
   top.  This, by default, is the same as before with first line being at
   the bottom.

MOT - 3/12/05
   Modified the WAIT command to include a timeout capability.  However, it
   doesn't work with XGenplot (no way to peek the keyboard buffer).  Needs
   more code to be done correctly.  But did get the function to return a
   code $key with the character pressed to terminate the wait.

MOT - 2/22/05
   Finally fixed and implemented properly the contour function.  Based in
   the plot function.  2D plotting of a surface now defaults to contours.
     read 9e.bmp -bitmap b1
     plot b1 -contour -dz 40 -dz2 5 -zscan 10 150 -rainbow -zrange 0 150
      -contour                             Contour map (see more options below)
         -dz <val> -dz2 <val>              Major and minor intervals on contours
         -zscan <zmin> <zmax>              Start and ending values for contours
         -zrigid                           (default) Draw only points from contour
         -zspline                          Fit contour points with spline for curve
         -zsmooth <strength>               Fit contour points with smoothing spline
         -rainbow [-zrange <zmin> <zmax>]  Use pretty colors, with defined range
     create -surface s1 x^2-sin(2*pi*y^2) -range -1 1 -1 1 -row 500 -col 500
     pl s1 -zrange 0 1 -rainbow

MOT - 2/22/05
   Added alias -export as command to print out all aliases in a form that
   can be directly added to a command file to regenerate.  Just convenience.

MOT - 1/19/05
   Alternate (and better way) to handle Ultratech problem is to add
   ability to specify the column delimiters in an ASCII file read.  This
   now exists with the -DELIMITERS "list" or -DELIMS "list" option.  To
   read based on & separating columns, use
      read test.dat -col 1 3 -delim "&"
   or for tabs
      read test.dat -col 1 3 -delim "\t"
   Not that it is not necessary to C-quote the string ("`\t") though the
   effect is the same.

MOT - 1/19/05
   Forced to add option "-noexpressions" to the READ command in ASCII
   format to deal with Ultratech :-(.  It prevents attempting to read
   expressions in a data file, returning error if there isn't a valid
   number in the column.

MOT - 1/19/05
   Added transform "VALUE_PAD" which pads to points with a specified value
         TRANSFORM VALUE_PAD <pts> <value>
   Very similar to ZERO_PAD except allows another value to be used.

MOT - 11/9/04
   Added ability to do a histogram on an array without first copying it
   to the main curve.  Also changed so that histogram takes all options
   that were created in the 2D histogram.
     transf hist -array wafer:z -dx 0.08
     transf hist -? works
   Old format is also still valid.

MOT - 10/21/04
   Added a 2D histogramming module.  Takes and X-Y curve and populates a
   2D histogram with frequency.  Surface will be created if necessary to
   hold the values.
     transf 2d_hist <surface> [-options]
   Lots of options.  Use transf 2D_Hist -? to get list.

MOT - 10/21/04
   Added option to TRANSFORM HISTOGRAM <dx> [-CENTER] which causes it to
   work like -CBIN

MOT - 10/18/04
   Added @stddev() as abbreviation for @std()
   Added @nearest(curve, x,y)
         @3d_nearest(curve, x,y,z)
   which return the index of the point "nearest" in distance from the 
   specified values.

MOT - 9/23/04
   Added a couple of functions for working with binary files and
   figuring out the format.  Mostly tired of writing C programs to
   determine if binary files are simple real numbers for some OS.
      int2hex(ival)          hex2int("string")
      float2hex(rval)        hex2float("string")
      real2hex(rval)         hex2real("string")
      double2hex(rval)       hex2double("string")
   Use if obvious.  Hex2<form> allow either 0xFFFF or just FFFF. 

MOT - 8/23/04
   Added ability to read in frames from an AVI movie (at least on
   windows).  Added more options in the read code.
     read <filename> -AVI <surface> [-FRAME <iframe>]
   If not specified, the first frame will be loaded from the file.
   When reading an AVI file, two internal variables will be set
        $FrameCount - number of frames in the file
        $FrameRate  - number of frames per second (usually 29.97)
    
MOT - 8/15/04
   Added function to start implementing continuum colors properly
      rainbow(x,xlow,xhigh)
   Returns a color (RGB) code corresponding to a continuum of colors from
   the current palette corresponding to the position of x within the range
   [xlow,xhig].  Can be used in any place where a color is requested.

   To set the palette, PLSYSTEM adds the command
      PALETTE RAINBOW [AFM, COLD, HOT, GREY, MOT]

MOT - 8/13/04
   Modified the read -FLIR code to handle three file types - 
      .fpf - public format that is well documented
      .fff - ir storage format.  Assume 320x240 and cal from Ultratech
      .seq - movie storage format.  Assume 320x240 and cal from Ultratech
   These are specificially to support my work and are probably not
   particularly useful elsewhere.

MOT - 8/11/04
   Added matrix feature to calculate the centroid of an object above a
   specified threshold.
       matrix centroid <surf> <level> [-silent]
   Calculates the centroid of a uniformly distributed object on the grid.
   Returns values as the linked variables
      X_C, Y_C   - coordinates of centroid based on X,Y coordinates
      Xi_C, Yi_C - index of centroid relative to the grid
      Area_C     - area (grid based) above the threshold
   The -silent suppresses the printing of the values.  If no points in the
   grid fall above the specified threshold, then all values are returned
   as 0.0.

   Added two additional options:
     -CM calculates the center of mass on the 2D grid.
     -CP calculates the power center - equal integrals on left/right
         only valid for integrals that are positive.

MOT - 6/28/04
   Added ability to read the "FLIR Public Image Format" for FLIR IR imaging
   cameras.  Use is probably limited to me, but so be it.
       read [-debug] -flir <surface> <filename>
   Similar in operation to the bitmap read capability.

MOT - 6/15/04
   Added the automated ability to read time-stamps from a data file.
   In the ASCII read, if the token contains a :, an attempt is made to
   interpret it as a time stamp in the format of hh:mm:ss or mm:ss.
   The component before the : can be one or two digits, but the rest
   must be exactly two (10:8:40 is invalid, but 8:40:20 or 08:40:20
   are equivalent).  The last element may include a decimal portion.
   Examples:  10:20:18.84

   Various checks are made for a valid time stamp.  Wraparound over a
   24 hour window is handled heurestically - should be okay as long as
   there is at least one element each hour.  Note that the value is
   ultimately converted to a REAL value and may lose precision.

   This feature is completely hidden and automatic.  Should just work
   as one would desire.
   
MOT - 6/7/04
   The generic "path search" algorithm now supports the automatic 
   recursive search using the ! and !! emtex nomenclature.
     macropath "adir;bdir!;cdir!!"
   will look for macro files in
       c:\adir   -- only in that directory
       c:\bdir!  -- in adir and all subdirectories immediately under it
       c:\cdir!! -- in bdir and recursively following entire tree
   Allows a little better organization of macros.

MOT - 6/7/04
   ^Y and ^V in XGenplot and XRump now paste the clipboard to the screen.
   ^Y was supposed to be used with cut/paste in line edit, but never has
   worked.  So make it equivalent to Emacs.  ^V is to suffer with Windows.

MOT - 6/7/04
   Modified the behavior PlotID so that the command
       identify "-5V bias" -nomark
   will properly work.  First string will be used as the ID even if it
   looks initially to be an option to the IDENTIFY command.

MOT - 6/7/04
   Added -Quiet as an alias in most (hopefully all) places for -SILENT.  As
   Q is a very unusual letter, the alias is accepted as -Q in code.  But as
   last tested, it may occasionally fail with just the single letter.

MOT - 5/5/04
   Added a new file read mode.
       READ -bitmap <surface_name> <filename>
   This new command just simplifies working with data files that are stored
   as bitmaps.  It interprets the bitmap as a surface matrix and returns the
   values.  Ignores all the color mappings, etc.  Each lookup index is
   treated as the actual data value (Z).  Limited to byte (8 bit) level
   bitmaps with are stored uncompressed.

MOT - 3/31/04
   Added new transformation
       TRANSF y d2y/dx2
   giving direct 2nd differential of the data.  In practice, taking the
   first derivative twice may be less noisy since uses 5-points versus
   only three in this algorithm

MOT - 3/15/04
   Added two new commands in the plot subsystem.
       CALC_AUTOScale    <rmin> <rmax>
       CALC_AUTOLogscale <rmin> <rmax>
   Determine the best values for an "axis" range encompassing these
   values.  Returns only internal variables for use:
       rmin$,rmax$ - recommended range
       dx$,dx2$    - recommended major and minor tick mark intervals
       mx          - labeling mode (# of decimals, or integer)
   Uses slightly different criteria for log and linear axes.  But
   same routine as used by axis routines to set GENPLOT versions.

MOT - 3/15/04
   Fixed VSORT under XP.  Now uses named pipes to communicate rather than
   attempting simple pipes.  Never found out why so hard to communicate
   between a Window application and a Console application, but so be it.
   Really should fold back into the code, but that is for another day.
   This could well break earlier versions of Windows - certainly
   Windows 95.  Don't know about 98, but may give a check.

MOT - 3/7/04
   Added
      MATRIX EXTRACT <old> <new> <r1,r2> <c1,c2>
   to extract into a new suface a limited subset of the first.  Copies
   rows [r1,r2] and columns [c1,c2] inclusive into new surface.  Alias
   is MATRIX SUBMatrix.

MOT - 3/4/04
   Modified and expanded the MATRIX command.  Moved the surface
   smoothing, rotating, and contour to the MATRIX default, but left
   in links via the old (well recent) TRANSFORM commands.  MATRIX is
   now the 3D surface equivalent of the TRANFORM command.  Added also
   the (now default) help MATRIX -?
      LIst | ?
      ABORT
      DUPlicate <old> <new>
      TRANspose <surf>
      ROW_get <surf> [<irow> | <r1,r2>]
      ROW_Put <surf> <irow>
      COL_get <surf> [<icol> | <c1,c2>]
      COL_Put <surf> <icol>
      TO_Curve <curf>
      TO_Surface <cv> <surf>
      ROTATE <surf> <deg>
      SMOOTH <surf>
      CONTOUR <surf> <z> [-sort]

MOT - 3/3/04
   Added MATRIX commands to put a specific col/row back into a surface
   matrix.  Very useful for rolling/shifting large data sets.
   Also fixed a minor but in MATRIX DUPLICATE which failed to copy all
   of the Y values on duplication.

   Added option TRANSF CONTOUR <surface> <level> [-nosort] which now
   returns only the crossing points without trying to sort into
   contiguous curves.

MOT - 2/21/04
   Yun/Shaoyin are becoming the next Pat Smith.  Limits on the possible
   number of columns in a data file has been increased from 1024 to 16384.
   The maximum "undefined" matrix size is now 16384x16384.  Defined
   matrices can still be any size.

MOT - 2/20/04
   Added a couple of transformations for working with 3D surfaces
       TRANSF CONTOUR <surface> <level>
   returns a sorted curve representing the contours at that level.
   Multiple isolated regions may be included in the curve, hopefully
   as curves one after another.  There are glitches for small areas,
   but these will be hopefully resolved.  X and Y returned values
   are the actual X and Y values from the surface, interpolated as
   necessary.

   Added 3D surface rotation
       TRANSF ROTATE -SURFace <surf> <angle>
   will rotate in place the surface by a clockwise angle.  The data are
   taken presently to be a uniform dx,dx grid, so pixel location rotated
   (this may be reconsidered soon).

MOT - 1/20/04
   Getting tired of some crashes caused by my tolerance to bad behavior.
   Added code to prohibit writing to array elements that are beyond the
   defined range.
       alloc a array 5 let a[5] = 7
   will now fail and print an error message.  This seems to be a major
   cause of failures.  To get back to the old behavior, MATHMODE now has
   a SAFETY OFF option to allow for bad behavior.  Anything goes once
   safeties have been turned off.

MOT - 1/20/04
   Added the help to MATHMODE command.  "mathmode /?" will access.

MOT - 1/20/04
   Added code so that 3D surfaces are read and written in binary mode
   preserving the column and row coordinate values.  This has always
   been true for ascii reads/writes, but binary defaulted to using the
   index number.  Implemented as an additional two columns - the first
   with the nrow values along the Y axis (same number of points as the
   actual data columns), and ncol values along the X axis.  New data
   file format of "GENPLOT VERSION 4.1" - used only for surfaces.

MOT - 4/22/03
   Added code so that 3D surface plots are only drawn over the range
   of X and Y on the axes.  Limiting the range will draw a subset of
   the surface as appropriate.

MOT - 2/14/03
   Kludge to allow breaking out of a control loop.  Only works with
   XGENPLOT and XRUMP.
   At initialization, a variable $break is defined by the key watching
   routine.  Everytime ^C is pressed, the value of this variable is
   incremented.  A rudimentary ability to escape out of a loop might be
      let $break = 0
      loop 10 {
         if ($break != 0) break
         ... other work ...
      }
   There ought to be a better way, but this works for the immediate
   problem.

MOT - 12/1/02
   Added asctime(-1) as an alias to ctime(-1).  Can't ever remember the
   ctime(-1) function but keep trying the asctime() version.  The correct
   C version would require a structure which we don't have - so isn't
   too much of a tacky move.

MOT - 12/1/02
   Added identify -unskip to backup one level on IDS.  Can be used as
   kludge for various activities - including multiple graphs per page.

MOT - 11/12/02
   Added // as a comment indicator.  Erases the end of the line.

MOT - 11/11/02
   Added internal variable to set default response to unwind operations
   due to errors.  Unwinding walks through blocks and macros asking whether
   they should be terminated or not.  In some automated operations, it is
   nicer just to have automated YES or NO response.
      $LexFlushMode - 0 ==> Normal query of how to handle unwind of macros
                      1 ==> Always answer YES to the queries
                      2 ==> Always answer NO to the queries
      $LexFlushCount - Incremented each time a FLUSH is handled with something
                       to do (NOP LexFlush() calls do not count).
   $LexFlushMode is reset to query mode on a RESET or RESET -FULL command.
   Should be used in macros where the behavior is clearly understood.

MOT - 11/11/02
   Added code to be able to automatically prime the response following a ZEN
   error message.  On fully automated macro runs, it is useful to suppress the
   the need to answer YES or NO to each continue request.  Internally, the
   routine LexFlush() is extended to LexFlushEx(mode) where mode is one of
   FLUSH_QUERY, FLUSH_YES and FLUSH_NO.  Yes and No give automatic responses
   instead of asking for how to handle the continuation of the macro/repeat.

   In addition, added an internally linked integer variable $SigTraps which
   stores the number of signal traps that have been detected.  This variable
   can be reset by the user to zero and is automatically incremented on each
   of the ZEN error type events.

	$ZEN_Traps - Number of ZEN errors that have occured.  Can be reset to 0.
	$ZEN_Reply - Default response after ZEN error.  0 => query
                                                        1 => answer yes always
                                                        2 => answer no  always
   Note that the ZEN behavior IS NOT ever reset by GENPLOT.

MOT - 10/23/02
   Added XGENPLOT as an alias for GENPLOT in GENPLOT.  So "xgenplot reset -full"
   works.  Just getting too used to XGenplot rather than Genplot.  Oops ...

MOT - 10/9/02
   Added option to CULL of IRANGE
     CULL KEEP IRANGE imin imax
   corresponds to XRANGE and YRANGE equivalents

MOT - 9/6/02
   Tired of fighting programs with poor understanding of parsing.  So
   GENPLOT default ASCII write is now tab delimited rather than space
   delimited.  Everyone using EXCEL should be happy.  Changed also from
   specifying %14.7g to just %g - tabs will force alignment instead.

MOT - 8/21/02
   cd and dcd commands now attempt to interpret their strings as expressions
   if the text itself does not resolve to a directory.  So
      declare mdir = "c:/Documents and Settings/tommy/Desktop"
      cd mdir
         -or-
      cd "f:/cgs"+"/"+"genplot"
   will now work.  String expressions are highly limited since spaces act
   as token separators (terminating a token) in LexGetFile code.  But use
   of "+" as a concatenation operator is very useful (as above)

MOT - 5/18/02
   Got annoyed with NLSFIT moving variables if you change the limits.  So
   now they stay in place on modification.  Helps to keep track of where
   the data is in the sigma$ variable.

MOT - 5/18/02
   Modified pwd() so that it translates backslashes to forward slashes.  All
   internal functions happily accept forward slashes in directory and file
   names - it's only the stupid command.com that couldn't accept the UNIX
   standard in the first place.  Since backslashes normally signify escaped
   characters in strings, it makes much more sense to have the default
   pathnames come out correct in labeling.
      title pwd()
   now works correctly.

MOT - 5/18/02
   Added bitwise logical operators.  Want to be able to make bit flags.
     a|b  => nint(a) | nint(b)
     a&b  => nint(a) & nint(b)
     ~a   => ~nint(a)
   This MAY potentially break some macros that use & or | alone for logical 
   and and logical or operations (instead of && and ||).  If there are enough,
   then I'll have to make a flag to enable.

MOT - 4/17/02
  Some commands with ON/OFF where YES/NO are equally reasonable may now
  tolerate either.  Such as LOGAXIS [YES } NO | ON | OFF]

MOT - 4/17/02
   Changed handling of multiline in ANNOTE.
   Two changes:
     (1) @end is searched for after whitespace.  So can be indented.
     (2) Look also for "sprintf(" after whitespace.  If there, is evaluated.
         Allows for arbitrary string expressions as text lines.

MOT - 3/18/02 
   Added new plot option
     pl -vecsize 0.001
   Changes the repeat vector length on dotted lines.  Same as the vecsize
   command but only affects this plot/overlay.

MOT - 3/18/02 
   Added a new plot option to enhance the lines and symbols mode.  
     pl -l&s -spoint 20
   This command draws lines and symbols.  The lines are drawn at full
   density while only every 20th symbol is drawn.  -spoint is similar
   to -npoint, but affects only the symbols.  -npoints affects either
   both, or just lines if -spoint is given.   Obviously, -spoint must
   be greater than -npoint (and is enforced).

MOT - 3/18/02
   No one is ever satisfied - I should really learn :-)  While many beat to
   get the full filename completion mode on TAB, others wrote to ask for the
   old behavior.  So have to give both :-)
        CONFIG TABCOMPLETION [ Full | Partial]
   The full will cycle through matching filenames, the partial will complete
   up to the first non-unique character.  Default is FULL.  You have to put
   it in GENPLOT.INI or RUMP.INI to switch by default.

MOT - 3/18/02
   Under UNIX, there is a problem in that pathnames correctly start with 
   the normal / character.  Due to DOS screwups in the early days, the /
   is an option identifier for PC's.  What a pain - GENPLOT/RUMP adopted
   this and now is stuck.  It is a serious problem especially for the 
   read and write commands which allow options on either side of the
   filename.

   The FIX: in read and write, only the - is accepted as an option 
   identifier.  The / is taken to be either invalid or the start of a real
   pathname.  I hope no one is adversely affected by this change.  If there
   are other commands where this should be the default as well, let me
   know and I'll change them.  It is now a trivial change.

MOT - 3/18/02 
   Added new option to fix_grid.  It now accepts also a -by <inc> option
   in lieu of setting the number of points.

MOT - 3/17/02
   Added cwd() as an alias for pwd() - both return the current working
   directory.
      declare current_dir = pwd()
        ...
      dcd %current_dir%

MOT - 3/8/02
   Changed the CHANGE DIRECTORY menu item on XGENPLOT and XRUMP to use
   the shell SHBrowseForFolder function.  Should be more stable compared
   to the kludge use of browsing earlier.

MOT - 3/5/02
   Modified history more.
      !<string>  ==> Retrieve last command starting with <string>
      !?<string> ==> Retrieve last command containing <string> anywhere
      !<n>       ==> Retrieve command <n>.  Use "history" to show list.
      !<-n>      ==> Retrieve nth previous command.  !-1 is last one.
      !!         ==> Retrieve previous command (any additional text ignored)
   Fixed so !! doesn't hang the system, except as the first command :-)
   ! commands are NOT stored on the history stack.  They simply do not
   change the history stack at all.

MOT - 2/20/02
   Modified the handling of the history retrieval commands.  If you use
   !<text>, this is the UNIX command completion command.  The previous stack
   of commands is scanned for one beginning with the specified text.  If you
   press <RETURN>, the command is simply executed.  This has always been the
   case.  But now, if you press <TAB>, the matching command line will be
   put on the line for editing.  Only the last matching line is given.
      !<text>  - find the last line with <text> as starting
      !<nn>    - retrieve the <nn> command line.
      history  - shows the history of command lines (to determine nn)
   TAB on the first two will complete the command line without executing.

   In addition, if the command processor interprets the !<text> command,
   it will print out the actual command line pulled from the history.

MOT - 2/12/02
   Fixed the dev -t (alias t) and dev -g (alias g) commands so they work
   with the newest changes forced by Microshaft in XP.

MOT - 2/11/02
   XGENPLOT and XRUMP now respond to key presses ^L and ^Z as follows:
     ^L - redraw both the text and graph screens if visible
     ^Z - minimize (iconize) both the text and graph screens

MOT - 2/7/02
   Added \Ohm, \ohm, \Mho and \mho as aliases to \Omega and 'S' respectively.

MOT - 2/4/02
   Added link to internal variable for the search path for macros.  Like
   the command, it is $BaseMacro.  You can set or modify it as you please.
      Basemacro c:\user\macros;c:\progra~\cgs\genplot\nt\local;
      let $basemacro = "c:\user\macros;c:\progra~\cgs\genplot\nt\local;"
      let $basemacro = "c:\user\macros;%$basemacro%"a
   As per Kevin request.

MOT - 2/4/02
   After much prodding, have changed keyboard control so the TAB works
   for filename completion as in Windows XP.  On the first press, it scans
   backwards for the start of a filename and then searches for all files
   that match the given text.  If there are no wildcards in the string, an
   implicit * is appended.  TAB then cycles through each of the files in
   turn until you press some other key.  <SHIFT>TAB walks backwards through
   the same files.  To see every file in the current directory, just type
   * and then TAB.  

   This code should work in UNIX as well - though it has not been tested.
   The <SHIFT>TAB works only in the XGenplot/XRump versions in windows
   and almost certainly not at all in UNIX.

MOT - 9/30/01
   Pat's been at it again.  fgets now permits up to 4096 characters.

MOT - 7/17/01
   Kludged the ability to get time/size info on files when specified as
   a UNC with //name/share instead of \\name\share.  (forward/back slash
   problem in windows).  Handles only filesize, filedate, isfile and isdir.

MOT - 7/17/01
   Pat's been at it again.  The sprintf (and hence printf and fprintf)
   functions were limited to 256 character output strings.  This has now
   been increased to arbitrary, subject to the limitation that printf
   specifiers other than  %s (ie. %f, %d, %p, etc.) are only GUARANTEED
   to have 2048 bytes available to them.  %s specifiers are guarenteed to
   have the larger of 2048 bytes or the length of the string to be printed.
   Also fixed EVAL to be able to print arbitrary long strongs.  THIS HAS
   ONLY BEEN TESTED TO 120,000 CHARACTERS :-)

MOT - 6/23/01
   Added DEV -REPAINT or DEV -REDRAW to force redraw of the graph screen

MOT - 5/8/01 - still long time
      Added reset option
           genplot reset -silent
      This just doesn't spit out the version info - nice for macros

MOT - 3/11/01 - been a long time since adding anything :-)
      New plot options
         PLot -SCOPE  - erases the existing screen and draws a "scope" face
                        with the data.  Cute, but not terribly functional.
         PLot -NPOint <n> - single use varient of the NPOINT command.  Causes
                            only every nth point to be used for drawing.  This
                            applies to both lines and symbols, but obviously
                            is useful primarily for symbols.
      New command
         SCOPE - draws a scope face.  You must use REGION commands to set the
                 range, or use the PLOT -SCOPE option.

MOT - 9/26/00
      Added the functions:
          sech(x)     csch(x)      coth(x)
         asech(x)    acsch(x)     acoth(x)
      Surprised that they were not already there.

MOT - 9/3/00
      Minor change in how dots in dash-dot line types are drawn in the
      Word Perfect driver.  Primarily to enable publication quality plots.

MOT - 6/26/00
      Changed "menu" read/save/save as routines so that they use the current
      working directory when first opened.

Speed-up: Windows 95/98 are absolutely abysmal with respect to the standard
          memory management functions malloc(), calloc(0, free(), etc.  These
          are supposed to be "free" functions in a modern operating system,
          but obviously no has bothered to tell the Microshaft engineers 
          that fact of life.  NT has very good algorithms, but 95/98 are
          an abomination before the Lord.  This severely impacts the 
          performance, especially on computation extensive operations like
          NLSFIT (lots of allocations in the thousands of dynamic evaluations
          required).  Since I am forced by the Microshit engineers to live
          with 98 on my laptop, the code has been kludged to bypass this
          restriction.  I manage the memory myself now for a factor of 4-6
          improvement (yes, not a percentage but huge factors).  The change
          under NT is negligible (but not any worse).  Please write and
          support the breakup of Microslave to multiple companies, 
          especially operating system companies that will truly compete.

Added: forlists command.  This is similar to foreach but the variables in the
       list are taken to be string arguments and expanded before the wildcard
       processing and list processing.
            declare alist = "a b c"
            declare blist = "c d e"
            forlist (alist blist) echo %f

Modified: The SOLVE now has an -noerrors options which prevents aborting when
          the root does not exist in the specified interval.  Useful mainly
          for macros that can deal with that condition otherwise.  Will return
          the limit with the smallest absolute error.

Added: Function @index(<array>,value)
       Returns index of first point in the array equal to or greater than
       the specified value.  If no values are above the given constant,
       -1 is returned.

Modified: The @min type functions now have an optional two arguments giving
          a beginning and ending index point for the scans.  The defaults
          are 0 to len-1, and given values will be limited within this
          range.
              @min(y)
              @min(y,npt/2,npt-1)  =  @min(y,npt/2,10000)
          The arguments are sorted so order does not matter.  Also, limited
          in range to [0,len-1] so can specify well beyond expected maximum
          for unknown length arrays.
          Affected functions:
                @min       @max       @sum       @ave      @avg     @mean
                @median    @variance  @var       @std      @rms     @skew
                @kurt      @range     @span      @absmin   @absmax  @abssum
                @absavg

Added:   Write -? is now implemented.  Also changed -NOTEXT option so it
         also turns off writing of the default curve identifier.

Added:   Multiple symbols on same line now enabled on IDENTIFY command
            identify "text" -MSYMbol 3 4 5 /

Added:   48 bit versions of random number generators.
            drand48()          ->  returns uniform [0,1] with 48 bit cycle
            lrand48()          ->  returns uniform [0,2^31-1] with 48 bit cycle
            mrand48()          ->  returns uniform [-2^31-1,+2^31-1]
            srand48(int seed)  ->  changes initial seed for all 48() routines
          The routines
            drand(), lrand() and mrand() are aliased to the above.
          In addition, the gnoise() function now uses this random number seed
          instead of the default rand() earlier provided.

          These routines will eventually replace rand() and rnd() [ie. they will
          become aliased to drand48()].

Added:    Alias in expression evaluator
              isvar("var")    <==>   exists("var")
          return 1 (TRUE) if the variable has been defined in the function
          evaluator.

Modified: The PM device driver now looks to see if you are running
          Windows NT or Windows 95.  If not NT, then it automatically
          does the bug fix kludge on printing and copying to the
          clipboard.  So the titles on the toolbar buttons are now
          correct for both NT and 95.  Don't know if Microsoft has
          fixed the bug in 98 yet.

Modified: The file opening functions in the function evaluator now
          trap and allow opening of the standard file handles.  This
          allows use of novel screen controls.  Use of stdin is 
          considered *VERY* dangerous and I don't warrent any use.
          It is okay under NT, but has not been tested otherwise.

          fopen(filename [, mode])
              filename == stdout   :  Open standard output (screen)
                          stdin    :  Open standard input (keyboard)
                          stderr   :  Open standard error (normally screen)
                          con      :  Open stdin or stdout depending on
                                      if mode contains 'r' or 'w'
                          /dev/tty : Identical to con

          Example: Outputting a countdown timer.  Reading in a character.
             setv stdout = fopen("stdout")
             setv stdin = fopen("stdin")

             loop %j=100,0,-1 { qev fprintf(stdout, "`%j\r") sleep .1 } 

             setv one_char = fgetc(stdin) eval one_char
             declare a_string = fgets(stdin)

          Caveat -- it isn't possible to close any of the standard handles.
          fclose() always returns 0 but the handle remains valid.  This is
          only true for the standard in/out/error.

Change: Precision of the error functions has been dramatically
        improved.  erf(x), erfc(x), ndtri(x) are now valid to 
        an fractional error on the order of 7E-10 over almost the
        full range.  These include the functions:
            erf, erfi
            erfc, erfci
            ndtr, ndtri

Added: Function for inverse error functions.  Similar to ndtri(x)
          erfi(x)  - inverse error function
          erfci(x) - inverse complementary error function
       Extended range for ndtri and validity to better match the 
       new ERFC range/precision.  ndtri(x) now valid
             [1E-307,1.0]  < 1E-8    absolute error
             [1E-307,1.0]  < 1E-8    relative error
 
Addition: FIT SPLINE has changed with addition of the -LINEAR option.
          Also, it is no longer necessary to do a sort before the spline
          fit call.  The data will be sorted if necessary by the spline
          routine.  However, the STRICT requirement remains (ie. the data
          cannot contain two points with the same X value).  Use 
          SORT -STRICT to eliminate duplicate points.
 
             [ FIT ] SPLINE {-options}

             Options:
                 -SMooth <RMS dev/pt>  - minimizes curvature by allowing
                             finite variation from the given knots
                 -SIlent   - Don't print smoothing info during processing
                 -LINear   - Piecewise linear, rather than cubic, spline
                             but otherwise using spline formalism.  Makes
                             it easy to use spline functions (dspln, etc.)
                             Ignored with -SMOOTH option

Addition: File control in function evaluator now includes pipe capability.
             setvar funit = popen("unzip -lqq zipfile", "r")
             qev pclose(funit)
          Usage:  FileHandle = popen("command string" [, "mode"])
                  int = pclose(FileHandle)

          popen() opens an external pipe using the specified command and
          directs the input/output from/to the file handle.  Normal fgets()
          and similar commands can be used to read/write to the pipe.

          Example:
              setv funit = popen("unzip -lqq aziz")
              while (feof(funit) == 0) fgets(funit)
              qev pclose(funit)
          A somewhat complex way to print out the contents of a zip file.

Clarification: Handling of escape sequences in character strings.

          There are several complications arising from the use of the
          backslash character in strings - which unfortunately will not
          generally go away.  The C language escape characters are very
          useful, and indeed necessary, for the printf functions.  But
          they make use of special strings and paths very difficult.
          (how to specify Microsoft paths, how to get \beta chars).

          So, there are now two types of string constants.  
             Regular string constants   - no escape constants
             C string constants         - full escape constant processing
          A string is taken to be a C string constant if it is enclosed
          in "" and the first character is a ` (the character may change
          if someone has a better idea - but the behavior described below
          will stay).  In C string mode, the \ character identifies an
          escape character to follow -- with the following recognized.
              \a  -> alert (bell sounds)
              \b  -> backspace
              \f  -> formfeed
              \n  -> newline
              \r  -> carriage return
              \t  -> horizontal tab
              \v  -> vertical tab
              \\  -> backslash
              \xbb -> hexadecimal constant where b is [0-f]
              \ooo -> octal constant where o is [0-7]
          Any other character returns just the character, so \w is 
          identical to w.

          To force a regular string beginning with the ` character, 
          use it twice.  This will disable the C string processing.
              "``0.834\beta"   gives the string  `0.834\beta
              "`\`0.834\beta"  gives the string  `0.345<bs>eta

          Eventually all strings will run through this processing.  At
          the moment, it is still a bit mixed.

          For example:
             strlen("`\n") = 1
          now returns 1 instead of 2 (or 3).  This is necessary to implement
          the fputs() function with line feeds etc.  I have resolved much of
          the trouble this may present elsewhere, but probably not all.  Let
          me know.

Addition: lnerfc(x) - returns ln(erfc(x)) valid over entire domain.  Uses
          alternate expansions near 0 and for large arguments.  This should
          provide uniform validity to 1E-7 fractional accuracy over full
          range.  Extended validity range is limited to real numbers.
          Complex is implemented as log(erfc(z)) directly.

WARNING: With the addition of the file handling capabilities, there was a
       major amount of change in the function evaluator.  Let me know if
       there are any unusual behaviors cropping up.

Addition: New silent-evaluation command 
                 qev   -> qevaluate     Quiet Evaluation
          Usage:
             qev fclose(funit)
             qev fputs(funit,"This is a line\n")
          Allows expression evaluation where the side-effect is desired
          without actually printing a value.  Identical to EVALUATE but
          will not print anything.

Change: Variable/Function defining routines now explicitly check to see that 
        the name specified is indeed valid.  Used to be able to do things like
            setvar 17 = pi
        although the variable "17" could never actually be accessed.  Now
        this will print an error.

MAJOR Addition: Significant file handling added via the expression evaluator.
        These will enable Pat to make much more complex macros to handle
        massive data manipulation.  The system provides the only limit on
        the number of files which may be simultaneously open.

  (1) New variable type: FILEPTR ==> a file handle.  Listed in LISTVAR
      and handled automatically by implicit setvar or SETVAR.

  (2) File read/write operations.   These are actually part of the math
      expression evaluator, but with implicit operations can be done at
      command level.
          stream = fopen("pathname" [, "mode"])       {implicit setvar}
              Opens the specified path.  The mode specifier is optional
              and defaults to "r" if not given.  Other values are
                    "r"   -> open for reading
                    "w"   -> create/truncate for writing
                    "a"   -> open for append write
                    "rb"  -> read binary   [probably not useful in GENPLOT]
                    "wb"  -> write binary  [probably not useful in GENPLOT]
                    "ab"  -> append binary [probably not useful in GENPLOT]
		    "r+"  -> open for read and write
                    "w+"  -> create/truncate for read and write
                    "a+"  -> create/append for read and write
		    "r+b" -> well you get the idea
		    "w+b" -> yep
		    "a+b" -> the same again
              The returned handle must be passed to all other file control
              functions.

          fclose(stream)                             {implicit int eval}
              Closes specified file handle, flushing all buffers.

          a_char = fgetc(stream)                     {implicit setvar}
              Returns next character from the stream.  Returns -1 if past
              the end of the file.

          a_line = fgets(stream)                     {implicit define}
              Returns next line from the stream.  The terminating newline
              character is removed from the string.  An empty line is 
              returned if past the end of the file (use feof(stream) to
              detect the end).

          fputc(stream, char)                        {implicit int eval}
              Writes a single character to the file (must be opened for
              writing).  Returns the character if successful, or -1 if
              unsuccessful.

          fputs(stream, string)                      {implicit int eval}
              Writes the string to the file (must be opened for writing).
              You must explicitly include the trailing "\n" if you want
              a line terminator.  Returns non-negative (usually 0) when
              successful and supposedly -1 if unsuccessful.

          fprintf(stream, format, arg1, arg2, ...)   {implicit int eval}
              Writes the formated string to the file (must be opened for
              writing).  You must explicitly include the trailing "\n" if
              you want a line terminator.  Returns the number of characters
              actually written to the file, or negative on fatal file
              errors.  The routine is tolerant, but abusive.

          a_line = sprintf(format, arg1, arg2, ...)  {implicit define}
              Returns the formated string as in fprintf.  Generally used
              with "define" command.  This following commands are 
              effectively synonymous:
                  s1 = sprintf("var %s = %g", "x", x)
                  sprintf s1 "var %s = %g" "x" x
              One uses the function evaluator, the other a GENPLOT 
              level command.

          feof(stream)                               {implicit int eval}
              Returns non-zero if end-of-file flag set on stream.
              Warning: A closed file handle will also return 0.

          ferror(stream)                             {implicit int eval}
	      Returns non-zero if error indicator is set on stream

          fflush(stream)                             {implicit int eval}
              Flushes all buffers associated with stream and returns zero
              if everything is okay

          ftell(stream)                              {implicit int eval}
              Returns an encoded form of the file position.  System
              dependent value.  May or may not be useful.

          fseek(stream, offset, mode}                {implicit int eval}
              Sets file position indicator based on offset and mode.  Mode
              is one of the constants SEEK_END, SEEK_SET, or SEEK_CUR which
              may vary between operating systems.  For NT, the values are:
                  SEEK_SET = 0
                  SEEK_CUR = 1
                  SEEK_END = 2
              Offset is based on the starting position above.  Returns zero
              if successful.  This is potentially very non-portable.


Change: The implicit EVAL on expressions now has a brief format look - just
        the result.  This may become the default for eval as well if I
        like it (and don't get too many complaints).

SCREWBALL: For UNC paths, apparently you cannot use // in place of \\
        in fopen.  Don't ask me why.  But
              x //lal/f/dieck/entr2.mac
        fails while any of
              x \\lal\f\dieck\entr2.mac
              x \\lal/f/dieck/entr2.mac
              x \/lal/f/dieck/entr2.mac
        succeed.  You figure it out.  I say crazy Microshaft.

Change: The REPEAT command has now been shown to fail with the advances in
        loop control offered by "loop" and "foreach".  So, until I hear of
        enormous complaints, the following command changes are implementd.
                REPEAT -> LOOP      now are synonymous
                OLDREPEAT           gets old version (if really needed)
                FOR    -> FOREACH   now are synonymous
                OLDFOR              gets old version.
        This was predicated when block control was created and has now
        happened.  The following gives very strange behavior.
                oldrepeat 20 times if (%c<5) echo %c
        Try it for curious results :-)

Add:    Aliases UNSETVar and UNSETLocal for dealloc and endlocal.

Add:    To accomodate the new RUMP spline fit feature, the function ispln(x1,x2)
        has been created.  This gives the integral of a spline fit (using
        SPL$DATA) over the specified range.
             ISPLN(xlow, xhigh)

Change: The unzip read option (using ::) now includes the -C option to match
        file names case insensitively.  Don't care if UNIX is case sensitive,
        GENPLOT will choose to be case rententative, but case insensitive.

Change: FFT routines are designed to work within powers of 2.  When the data
        set is other than a power of 2, there are two options for handling
        the problem.  One is to truncate to the next lower power of 2 -- the
        other is to somehow pad out to the next higher power of 2.  Truncating
        loses data -- padding artificially adds non-real values.  Neither is
        good.

        So -- give the user the option.  Three new FFT options.
            -TRUNCATE - truncate data to nearest 2^m below npt
            -PAD      - expand data to nearest 2^m above npt, filling in 
                        with the data set (copy from beginning)
            -ZEROPAD  - expand data to nearest 2^m above npt, filling in 
                        new data points as zero

        The default is -PAD which is a strange compromise.  Versions prior
        to 12/98 essentially mixed -TRUNCATE and -PAD depending on NPTMAX.
        But now the curve will be expanded as necessary to allow either
        -PAD or -ZEROPAD in all circumstances (up to 2^20 points/curve).

        AUTOCORRELATE is always ZEROPAD'd.  This makes sense.

        FILTER is constant extension padding (last value just used).
        This eliminates a discontinuity in the data, but leaves the
        discontinuity in the derivative.  Oh well ...
          

Fix: With expansion of loops, etc., problems have arisen in variables becoming
     inaccessible in nested loops.  The new code will allow up to 24 %x type
     variables properly nested and used -- handling reuse as required.
     
     Loop control statements (foreach, loop, while) always set the substitution
     variables %i (index) and %c (count).  This remains.  The foreach() and
     loop() may also set an additional variable of arbitrary letter.  These
     letter'd variables are now maintained as substitutions deep into loop
     blocks (but not across macro calls).  %%x as usual is converted into
     %x.

     Examples are the only way to properly demonstrate:

     Simple nested loop.  Notice innermost with replacement of %c.
        loop %a=1,2 {
          loop %b=3,4 {
            loop %%c=5,6 {
            echo %%a=%a %%b=%b %%c=%c    %%i=%i
        } } }

        %a=001 %b=003 %c=005    %i=000
        %a=001 %b=003 %c=006    %i=001
        %a=001 %b=004 %c=005    %i=000
        %a=001 %b=004 %c=006    %i=001
        %a=002 %b=003 %c=005    %i=000
        %a=002 %b=003 %c=006    %i=001
        %a=002 %b=004 %c=005    %i=000
        %a=002 %b=004 %c=006    %i=001
    
     Over-riding outer loop entries
        loop %a=1,2 {
          loop %b=3,4 {
            echo %%a=%a %%b=%b
            loop %%a=5,6 {
            echo %%a=%a %%b=%b    %%c=%c %%i=%i
        } } }
        %a=001 %b=003
        %a=005 %b=003    %c=001 %i=000
        %a=006 %b=003    %c=002 %i=001
        %a=001 %b=004
        %a=005 %b=004    %c=001 %i=000
        %a=006 %b=004    %c=002 %i=001
        %a=002 %b=003
        %a=005 %b=003    %c=001 %i=000
        %a=006 %b=003    %c=002 %i=001
        %a=002 %b=004
        %a=005 %b=004    %c=001 %i=000
        %a=006 %b=004    %c=002 %i=001

     Combination of loops and foreach block, with a while.  Again
     note use of %%c so it isn't replaced by %c from the %k loop.
        loop %k=1,2 {
           while (%k+%%c<4) {
              foreach %d in (mm, nn, zz) {
                 echo %%k=%k  %%d=%d  %%i=%i
        } } }
        %k=001  %d=mm  %i=000
        %k=001  %d=nn  %i=001
        %k=001  %d=zz  %i=002
        %k=001  %d=mm  %i=000
        %k=001  %d=nn  %i=001
        %k=001  %d=zz  %i=002
        %k=002  %d=mm  %i=000
        %k=002  %d=nn  %i=001
        %k=002  %d=zz  %i=002


Add: SETPRIORITY command (in system).  This command under NT modifies the
     priority level of the main computation thread from -2 to +2.  Default
     is +2 at the moment.  0 indicates default thread priority from the
     operating system.  Other values are relative with +2 locking the 
     system for compute bound tasks, and -2 becoming essentially idle.

     Requesting SETPRIORITY with an out-of-range value will print the
     current priority level.

     The default priority of the XGENPLOT/XRUMP process is set by the
     value of PROGRAM_PRIORITY=2 in the genplot_.ini file (normally found
     in "Program Files\cgs\genplot\nt".  This ASCII file can be edited to
     set the default priority to desired value.  Or SETPRIORITY can be
     added to the genplot.ini initialization file.

Add: Since NT allows the format "cd /d f:\cgs" changing disk and path, this
     is also now allowed for GENPLOT (NT and OS/2 only)

Add: Another string function
         fullpath( <partial_path> )
         Example: eval fullpath("./test")
     The actual file or path need not really exist -- it will be returned
     as closely as possible.

Minor: Allow || and && as logical or and and respectively.

Minor: colormap define now allows <ESC> to abort out of entries.

WARNING: There are two known instabilities at the moment.  With no activity,
         GENPLOT may simply hang.  Selecting it in the task bar and 
         executing CLOSE from the menu will properly terminate it.  There
         is a more serious hang that occurs with extensive usage -- seems
         to be related to font corrpution in the GENPLOT window.  This is
         being searched for now -- you may see occasional debug files show
         up on your hard disk from this testing.

New feature: Under HELP menu items, there is now a "Modify Serial Number"
     option that allows the serial number to be set from within GENPLOT.
     This can be used to register a valid number without reloading the
     program.  Also, once a valid serial number is entered, the program 
     remembers the true serial number.  On updates, you can enter anything
     as the serial number -- the program will automatically correct it
     when first run.

New functions: The value of atoi("junk") is 0 -- hard to identify an error.
     Added new functions isatoi(str) and isatof(str) which return TRUE if
     the ENTIRE string can be properly interpreted as a integer or floating
     point value, and FALSE if any character is invalid.  Also added function
     to do arbitrary base conversion - strtol.  This acts like the C function
     except it does not save the location of the first invalid character.
          isatof(string)
          isatoi(string)
          strtol(string, base)

Problems: Chebyshev polynomials do not work (in real mode) for OS/2 due to
          a compiler bug.  Since OS/2 is no longer long term viable, it is
          not a big enough deal to try to fix.

Modify: Override of the search path - variable is now system dependent
            GenplotSearchPath_OS2
            GenplotSearchPath_NT
            GenplotSearchPath_UNIX
        This environment variable will override the search path for
        system files.

Add:  New function dpoly() to calculate derivatives and integrals of 
      polynomials.
              dpoly(x,iorder,array)
        iorder => 0 is evaluate function
                  1 is first derivative
                  2 is second derivative
                 -1 is first integral (all constants of integration = 0)
                 -2 is second integral
      Derivatives of all order are allowed; integrals are only accepted
      to tenth order (makes no sense anyway).

Fix/Add: Modified the CCIRCLE command in ANNOTE to accept the radius as
         a cursor entry (any radius point).  There is some subtlety to
         use of user mode coordinates.  In either USER or INCH mode, the
         radius is still in terms of inches.  It will be properly handled
         and reported if entered by cursor.

         Added: PCIRCLE command
               PCIRCLE [x y radius | /] <start_degrees> <end_degrees>
         Draws partial arc specified by the start and ending angles.
         Angles are measured counterclockwise from the positive X axis.

Fix: The options to annote commands now default back to expecting a number
     if there was any problem in interpreting as an option token.  This
     allows negative values in expressions.

Minor change: The command GENPLOT, ANNOTATE, RUMP now have an trailing
        color (:) in their definitions.  Thus you can paste the output of
        a GENPLOT session directly back in without stripping the prompt.
        For RUMP, RUMP: will not print the definition.

Size change: The MATRIX read code no longer places any limits on the length
        of an ASCII line.  Paul Carey and Pat Smith are free to write 50,000
        character length lines as desired.  If this works, will be moved to
        be more generally implemented on all reads.

Document: Internal documentation of the PLOT and OVERLAY commands.  Added
        the PL -? and OV -? commands to give brief description of the myriad
        of options.  Should be done for many more commands.

Modify: Modified the edge (first/last point) handling of the histogram and
        bar graph modes.  Now tries to make last bins symmetric with respect
        to the data points.

Enhance: Added options to many of the annote commands.  The following will
      work with circle/arc/ etc. commands.
          -pen <colorname>
          -lw    | -linewidth  <linewidth>
          -ltype | -linetype   <linetype>
      Unlike many options, these immediately follow the command since the
      number and types of arguments may vary for many of the commands (for
      example the DRAW CONNECTED command)

Modify: The printf "%s" string operations will now permit full string
      expressions rather than just string variables.  Works with &encode
      and printf sequences.
            printf "%s" strcat("The time is now", ctime(-1))

Modification:  Substition of control sequences.  Advanced operation only.
      The substition for control sequences like %i, %c and %f in foreach,
      loop, while, etc. was previously done at "line read" time.  It has now
      been moved later to "token parse" time -- ie. just when the object is
      needed.  This may make a few complex macros malfunction, but makes it
      much more consistent in behavior.  Under the old sequence, the following
      two constructs behaved differently.  Under the new, they are equivalent.
           loop 5 { loop 5 { echo %c } }
      versus
           loop 5 { 
              loop 5 { 
                  echo %c
              }
           }
      In the first case, it used to print out 1,1,1,1,1, ...  5,5,5,5,5,
      while the second when properly 1,2,3,4,5,1,2,3,4,5,...  Both now
      give the correct sequence.

      Because this change required low level modification of the parsing
      algorithm, there is a high potential for unanticipated side-effects.
      Please let me know of any unusual behavior that may be related to
      this change (should only happen with %f, %i, %c and %var% type
      substutions.
       
Added command processor:
        pwd()  - returns current working directory
        getenv("var") - returns value of the environment variable
      Note the "declare name = $USER" works, but only for uppercase
      environment variable names.  "declare name = getenv("user") will
      work for any names.
      
Debug feature: The command "echo" is now expanded to include simplification
      for debug modes.  The command syntax is:

      Usage: echo [-? | ON | OFF | DEBUG | NORMAL | <message>

      ON and OFF set global echoing of command file lines.  DEBUG overrides
      echo and turns it on for all subsequent macro levels, independent of
      any other echo commands, until a ECHO NORMAL is executed.  Echo can
      also be used to print a single line of text.

Minor change: Removed the "sh", "os2" and "os/2" as commands for starting
      an external shell.  There still remain the equivalent commands
      "!", "dos", "ok;", "csh", and "shell".  sh was conflicting with
      show inside RUMP.

Minor change: Back by popular demand -- the ability to reverse the on-screen
      pm behavior from black lines on white to white lines on black.  An
      option in the SETUP dialog box of the PM screen allows you to choose
      whether pens 0 and 1 are reversed on the display screen.  Remember to
      "SAVE OPTIONS" after changing any settings.

      Note that running with reversed pen 0 and 1 violates the WYSIWYG mode
      of screen printing.  You cannot reverse the function of the pens for
      hard copy except by using PALETTE MAP commands.

Minor change: The file/directory test functions have been added to the 
      standard string operations.  This means that their actual use has
      changed.  
           isfile(pathname)   returns 1 if path is an existing file
           isdir(pathname)    returns 1 if path is an existing directory
           filedate(pathname) both return seconds since epoch, or -1 if
           filetime(pathname)   the file does not exist.
           filesize(pathname) size of file in bytes, or -1 if doesn't exist
      The pathnames are now generalized string expressions or variables.  If
      a simple filename is given, it must be enclosed in quotes.
           eval filesize("test.dat")
           declare basename "test"
           eval filesize(strcat(basename, ".dat"))

Minor Change: The handling of loop/list substitution has been changed with
        regard to inner block structures.  Conditional execution blocks now
        will have variable substitution from the controlling foreach, loop,
        or while block.  The behavior of conditional blocks within for
        and repeat control loop has not changed.
            loop 20 {
               if (%c < 5) {
                  printf "%d" %c
                  if (%c < 4) printf "Inner loop too: %d" 100*%c
               } else {
                  printf "%d" 2*%c
               }
            }
        Prior to this change, the inside printf would fail since %c was not
        substituted with the loop count.  Now, they behave as expected.

        WARNING: FOR %f and REPEAT n commands are maintained currently for
           backward compatibility and in case foreach, loop fail.  They will
           ultimately be eliminated and directed to the corresponding FOREACH
           and LOOP commands (expanded versions of the same).  DO NOT DEPEND ON
           THE ABSENCE OF VARIABLE SUBSTITUTION IN THESE COMMANDS!  IT WILL
           CHANGE SOMEDAY.        

Document: The area,page,axes fill are now implemented for the DVIPM and
        Postscript drivers.  Others will never see the light of day.  In
        encapsulated postscript mode, the page fill attempts to only go
        just beyond the actual drawn area.  This is a very kludgy and
        difficult mode to support -- recommend not using AREAFILL in
        encapsulated postscript unless you find that it really does work
        as you need.

Change: To properly handle (ie make it look good by default) the new
        area fill command, the margin on Y had to be increased.  For this,
        the new default SIZE is changed to 7.45x9.00 (from 7.30 by 9.00),
        the margin on the left is increased from 0.85 to 1.00 and the offset
        along X is decreased to 0.35 (from 0.50).  The net effect is that
        default graphs are exactly the same size in the same place.

Add: Added new commands
           PAGEFILLCOLOR
           AREAFILLCOLOR
           AXESFILLCOLOR
     These are decreasing area fills.  Page covers entire page (a rectangle
     sufficiently large to cover all of the page -- but boundaries of the 
     page are not included in the size for encapsulated postscript.  Area
     includes axes and the margins around axes.  And axes is internal area
     only.  Page flood is done on each erase.  Area and axes flood fills 
     are done on each axis.

Fix: Tried to modify code so "anno label 1 1 'text'" works without giving
     extra single quote marks.  In future, all strings should be enclosed
     in double quotes only -- but cannot enforce and may not be good to
     enforce.  In "", is evaluated as a string expression.  in '', simply
     taken as given.

Add: ^L (Control L) will now do a force redraw of the graph screen (if
     typed while the graph screen has focus).  This is a full redraw --
     also passes the ^L to the text window which will also do a full redraw.

Obsolete: The command COLORMAP has been superceded by the PALETTE MAP
          command.

Clarification: The number of pens used for autochange mode depends on several
         parameters.  The palette command specifies that a number of the
         entries are available as pen colors.  But the DEVICES.DAT has a 
         similar parameter.  The smaller of the two will be used.  So if a
         pen plotter has only 8 pens, then even if the palette is said to 
         have 20, only 8 will be used before recycling back.

New Command: PALETTE - in general graphics support.
         See example file \Program Files\cgs\genplot\NT\default.pal

         PALETTE DEFINE <num_entries> <num_pens> -- defines the palette
               o  num_entries is the number of entries in the palette.
               o  num_pens is the number of entries used as pens (not including
                     pen 0) for the autochange function.  Normally, it is one 
                     less than num_entries.  However, additional palette entries
                     may be defined for use as structural colors (axesfill,
                     etc.)

         PALETTE SHOW | LIST  -- shows current palette information.  Includes
               the status of the axes and page fill colors.

         PALETTE MAP <pen> <color> [<pen> <color> ...] /
               Identical to the old COLORMAP commmand.  Sets the individual
               entries in the current palette to the given values.

         PALETTE COLORLIST    -- prints a full list of all known colors

         PALETTE SAVE <filename> -- writes commands to recreate the palette
               to a .pal file.  Always written as a local file unless path
               given.  Sets the AXES and PAGE fill colors, as well as the
               palette.

         PALETTE LOAD <filename> -- loads a palette file
         PALETTE READ <filename> -- loads a palette file
               These are synonymous commands.  File is searched for in the
               current directory and along the system search path.  This
               command causes the contents of the file to be executed (as if
               typing "XEQ <filename>") with echo off.  

     /* ===========================================================================
     /* This file is a palette definition file.  
     /* 
     /* Blank lines or lines beginning with /* will be ignored.
     /* 
     /* The following keywords are recognized:
     /*   PAPERCOLOR     <colorname or RGB>
     /*   AXESFILLCOLOR  <colorname or RGB>
     /*   PALETTE DEFINE <count> <num_pens>
     /*
     /* PALETTE DEFINE:  Followed by two integers specifying the number of 
     /*           entries in the colormap and how many (not including 0) will 
     /*           be used as normal pens for autocolor changes.  The default 
     /*           palette is 16 colors (pen 0 through 15), although any size 
     /*           palette is supported (device drivers may have problems however).
     /*
     /* PAGEFILLCOLOR: RGB specification for the flood fill over entire paper.
     /*                Default value is "none".  Can be name, expression, or an
     /*                integer refering to the palette.  Page fill is only done
     /*                after an erase (autoerase or otherwise)
     /*
     /* AREAFILLCOLOR: RGB specification for the area fill color.  The area is
     /*                defined as the region including the internal axes area
     /*                and the margins defined for labels.  Will be drawn for
     /*                each axis generated.  Default value is "none".  Can be
     /*                name, or an integer refering to the palette.
     /*
     /* AXESFILLCOLOR: RGB specification for inside of axes.  Default value is
     /*                "none".  Can be name or integer refering to the palette.
     /* ===========================================================================
     PAGEFILLCOLOR none
     AREAFILLCOLOR none
     AXESFILLCOLOR none
     PALETTE DEFINE 16 3
       white
       black
       grey50
       grey100
       grey20
       grey30
       grey40
       rgb(50, 50, 50)
       grey60
       grey70
       gray80
       gray90
       darkcyan
       darkyellow
       tan
       blue

Color Handling:  Color inside GENPLOT/RUMP is now internally full RGB
         capable (but some commands still accept only pen numbers).  There
         is a new database file providing mapping of color names to RGB
         values called "colormap.dat" which resides in the usual place
         (Program Files\CGS\Genplot\NT\colormap.dat).  The file is self
         documenting.   The commands COLORMAP, PEN, AXESFILL, PAGECOLOR
         are certainly known to work with it.  As others are identified,
         they will be updated as well.

         The color database currently has 700+ colors.  The names are
         case insensitive but must be matched fully.  GRAYxx where xx
         is a two digit % are included (Grey7, Grey70, etc.)

         Only the PMDRIV for NT and the Postscript driver can actually
         make use of the full RGB capability.  Other drivers use the
         closest pen number to a specified color (pen numbers based on
         the current color map palette).  The number of colors is
         unlimited in the code, but may be limited by the actual 
         rendering of the device.  NT running in 24-bit color mode
         handles all of the colors well.

           axesfill CarnationPink axis
           axesfill RGB(28,448,330) axis
        
         New colors can be defined as a normal Genplot variable and
         used as in any expression.  Color matching first checks the
         color database and then interprets values as an integer expression
         becoming either a pen number or a color specification.  However,
         RGB colors should be saved as INTEGERS to utilize the full color
         space.  If not, the red component will be truncated to even values.

           alloc MyYellow INT  let MyYellow = RGB(220,190,60)
           axesfill MyYellow axis
           colormap 3 MyYellow /
           axesfill 3 axis
         
         Setting some colors cannot be done by name (such as internal
         variable $axcolr[]).  For these, a new function is implemented
         which converts a string to the RGB equivalent from the database.
           eval rgb_color("Yellow")
                16842751 = rgb_color("Yellow")
           printf "%8.8x" rgb_color("Yellow")
                0100ffff 
           let $axcolr[1] = rgb_color("gray80")
         (RGB specifications are 0x01000000 or'd with the 3 byte RGB)
            
         If you really want to get a full listing of the available colors,
         use the command "PALETTE COLORLIST"

         The palette may also be queried by directly examining the RGB
         entries in $PALETTE array.

COLORMAP: The colormap command changes the color mapping in the currently
          loaded colormap.  The version of the palette on disk is never
          changed.  For most devices, changing the colormap is not effective
          until the device is re-initialized.  But for the NT PM-driver,
          the changes are immediate on screen (existing plot is changed
          for all object drawn with pen specifications -- as opposed to
          specific RGB specifications).


Change: RGB(x,y,z) returns a value which you are not to understand.  But
        can be passed to PEN to set specific colors.  Documentation below
        changed.

Added: Color commands can now be specified as full RGB values.  A longer
       list of colors will be added soon.  Because of the change to a
       display map of colors, the 

Added: AxesFillColor (or AxisFillColor) - 8 character minimum
       Sets color for fill of axes in graph.  For subplots, might want to
       use 0 to block out other components.  Can also be various colors
       to standout on graphs.

Added: functions 
             atof("string")
             atoi("string")
             atol("strong")
       ascii to float, integer and long.  The second two are identical.
       The string is interpreted as a number as far as is consistent with
       the type of variable.  No error checking or return.

Added: Recognition of hexadecimal constants.  0x123 is properly interpreted
       as the integer 256+32+3 = 291.  Also the function RGB(r,g,b) which
       just packs the three values into a single 32 bit integer.  Fortunately
       a float has 24 bits of precision so can be represented exactly.
       NOTE: RGB is no longer specified.  The specific value is lost, but
       still represents a color.  Bottom three bytes are rgb with R in
       the lowest order position.

       RGB(r,g,b) -- packs three integers r,g,b into a 24 bit
          integer representing colors (RGB model).  The values should be
          between 0<=r<=255 -- actual operation truncates values to integer,
          and then just uses the values modulo-256.  So RGB(257,3,-1) is
          equivalent to RGB(1,3,255) -- or essentially blue.

Added: With the string expression evaluation, it is time to start migrating
       some of the commands to expect such expressions.  First two are
       TITLE and ANNOTE LABEL.  These are generally the most complex used.
       As I see fit, others will migrate as well.
            anno label / strcat("sample ",ids)
       Identification will probably move also, but need to think about
       at what level -- ie. can the IDS be an expression which is expanded
       just before being drawing?

Clarification: There are some real subleties added with string evaluation now.
          Consider:
             create y = sin(x) archive c1
             decl m = c1
          What should m be?  "c1", or the value of the IDS on curve c1 which
          should be "New Curve".  This is a problem!  I've solved it by
          scanning a string expression first to determine if it is simple or
          complex.  A complex one is any involving characers ({[+/#.  These
          signify that some sort of operation may occur.
             CASE: COMPLEX EXPRESSION
                 interpret as possible string variables.  Variables will resolve
                 to their ID if 2D/3D curve or to their definition if a function.
                 If the string c1 is required, enter as true string constant "c1".
             CASE: SIMPLE EXPRESSION
                 Even if name is a function or 2D/3D curve, treat is as simple
                 text and return only the text.  To get the value of a simple
                 string, use either #c1 (forcing complex) or %c1% (expression
                 expansion in the parser.
          Note that this DOES NOT apply to string variables which will always
          resolve to their definitions.  Use quoted strings to avoid this 
          behavior in a DECLARE operation.
             declare m1 = "c1"      /* Always safe
             declare m2 = "m1"      /* Always safe
             declare m3 = m1        /* Will set to value of m1, not "m1"

Addition: Added ability to specify string arguments in defined functions, as
          well as string returning functions.  The prepend character '#' acts
          as a prototype indicating that the variable (or function) name is
          to be considered a string type variable (or return value).  Different
          symbols may later be added for other type arguments.
              define f(#var,n) = strlen(copies(var,3))
              define #g(#var,i,j) = subword(var,i,j)
          Note that the # sign does not appear in the expansion, only in the
          function name.
              GENPLOT: define #g(#var,i,j) = subword(var,i,j)
              GENPLOT: define f(#var,n) = strlen(copies(var,n))
              GENPLOT: eval g("this is a long line", 1,3)
              Value: "is a long" = g("this is a long line", 1,3)
              GENPLOT: eval f(g("this is a long line", 1,3) , 3)
              Value:              27 = f(g("this is a long line", 1,3) , 3)
              GENPLOT:
          Now to see how much broke in adding this feature.

          If a string argument is encountered where a numeric value is expected,
          the string is interpreted as a simple number (no math operations).  In
          complex mode, it can be of the form A+Bj.  
              define f(#a) = 2*a
              eval f("34") = 68
              eval f("34+8") = warning messages and 68 as the value

Addition: Enabled "Filled Rectangle" drawing in the postscript driver.  The
          ANNOTE FILLED_R command will now properly output to postscript
          drivers.

Slight fix: There has been an annoyance that you had to re-enter the serial
            number everytime an update was loaded from the WEB.  The serial
            number handling code has now been revised to keep track of valid
            serial numbers that have been entered, and replace the "serial"
            entry if necessary.

            Functional: For updates, if you have once entered a fully valid
              serial number (WGEN-97-xxx-xxx), you don't have to enter it
              again.  You can enter anything for the serial number during the
              install process.  The correct value will be inserted first time
              the program is run.

Document: Use of the external macro facility in GENPLOT: - an Example
     go.mac
     ---------------
          /* ===========================================================================
          /* This is a dummy demonstration of the macro calling ability of 
          /* the nlsfit function.  The macro feature is included to allow fitting
          /* functions which cannot be defined as a simple f(x) type expression,
          /* for example those involving integral expressions.  Any necessary
          /* work can be done in the macro, including calling an external program
          /* and reading in the result from a temporary data file.
          /*
          /* In this one, we are just fitting a gaussian peak with a very complicated
          /* mechanism.  
          /* ===========================================================================
          
          /* Create the test data with some noise
          create y = 1.3*gauss(x,2.0,0.5)+0.2*ndtri(rnd(x)) -range -2 +5 -points 50
          
          /* Set nominal coefficients (see test_fnc.mac for use of these)
          setv a,x0,dx = 1.2 1.8 0.3    /* Modify so not exactly the same
          
          /* Allocate an array for test_fnc.mac to save its estimates to
          alloc results array npt      /* Array holding results from test.mac
          
          /* Enter NLSFIT and go - only addition is the genplot_macro feature
          nlsfit
            reset
            equ results                /* Will pick up the array created above
            vary x0 / 
            vary dx /
            vary a /
            genplot_macro test_fnc.mac
            fit
            return
          
          /* Plot the data and overlay the points estimated as the best fit
          plot
          let y = results ov -pen 2 -lt 1

     test_fnc.mac
     ---------------
          /* ===========================================================================
          /* This macro must evaluate the fitting function at each of the points
          /* in the data set.  It should assign to results[%i] the estimate of the
          /* function at the each of the coordiates x[%i]
          /* 
          /* Requirements:
          /*  (1) results[] must be filled with the estimated Y at each X
          /*  (2) the original data passed must not be changed (use archive/retrieve
          /*      bracketing around the macro to ensure this)
          /*  (3) the final line should be a "return" to exit the sub-process
          /*
          /* This test version does very little - any real use would have a much
          /* more extensive calculation for each of the elements.
          /* ===========================================================================
          
          archive hold_data
          loop %k=0,npt-1 {
            let results[%k] = a*gauss(x[%k],x0,dx)
          }
          retrieve hold_data
          return


Modify:  The is once again symmetry in the function evaluator between the
         string and numeric expressions.  'a' and "a" are equivalent when
         expecting a numeric argument.  'this' and "this" are equivalent when
         expecting a string argument.  Only "this" fails to parse when looking
         for a numeric argument.  This should make life easier on users of
         the extensive string functions.

Modify/Fix: The strip() function now has two optional arguments rather than
            one.  The format is:
              strip("string" [, "mode" [, char]] )
                  mode - "Leading" | "Trailing" | "Both".  Default = BOTH
                  char - must a single character to strip.  Defaults  ' '.

ADD:     More string functions - time related
              time()        - returns number of seconds since the Epoch
              clock()       - returns number of clock ticks program or OS start
              ctime(time)   - converts a time (in seconds since epoch) to an
                              ASCII string with the default format 
                              Sun Dec 02 06:55:15 1979\0  (trailing \n stripped)
                              If time is specified as -1, then current time is
                              used - equivalent to ctime(time())
              strftime(format, time) - encodes time into a string based on the
                              C equivalent function strftime.  See any C
                              reference text for the parameters in the format.
                              If time is specified as -1, then current time is
                              used - equivalent to strftime(format,time())
         In the conversion to C, these are implemented as:
           ctime(time)             ->  ctime(time)
           strftime(format, time)  ->  strftime(buf,256, format, localtime(time));
         
         GENPLOT: eval strftime("%X %a %b %d", -1)
         Value: "08:19:20 Fri Feb 20" = strftime("%X %a %b %d", -1)
         GENPLOT: eval ctime(-1)
         Value: "Fri Feb 20 08:19:28 1998" = ctime(-1)
         GENPLOT: eval ctime(time()-8*24*3600)
         Value: "Thu Feb 12 08:19:47 1998" = ctime(time()-8*24*3600)
         GENPLOT:

           %a Abbreviated weekday name
           %A Full weekday name
           %b Abbreviated month name
           %B Full month name
           %c Date and time representation appropriate for locale
           %d Day of month as decimal number (01 – 31)
           %H Hour in 24-hour format (00 – 23)
           %I Hour in 12-hour format (01 – 12)
           %j Day of year as decimal number (001 – 366)
           %m Month as decimal number (01 – 12)
           %M Minute as decimal number (00 – 59)
           %p Current locale’s A.M./P.M. indicator for 12-hour clock
           %S Second as decimal number (00 – 59)
           %U Week of year as decimal number, with Sunday as first day of week (00 – 53)
           %w Weekday as decimal number (0 – 6; Sunday is 0)
           %W Week of year as decimal number, with Monday as first day of week (00 – 53)
           %x Date representation for current locale
           %X Time representation for current locale
           %y Year without century, as decimal number (00 – 99)
           %Y Year with century, as decimal number
           %z, %Z Time-zone name or abbreviation; no characters if time zone is unknown
           %% Percent sign

           Format Code Meaning 
             %#a, %#A, %#b, %#B, %#p, %#X, %#z, %#Z, %#% # flag is ignored. 
             %#c Long date and time representation, appropriate for current locale. For example: “Tuesday, March 14, 1995, 12:41:29”. 
             %#x Long date representation, appropriate to current locale. For example: “Tuesday, March 14, 1995”. 
             %#d, %#H, %#I, %#j, %#m, %#M, %#S, %#U, %#w, %#W, %#y, %#Y Remove leading zeros (if any). 

MODIFY:  Handling of comments in 3D Matrix read:
         Within the header section (before the data), the following are
         allowed:
              C             <== comment which, if first, becomes IDS
              #             <== comment which, if first, becomes IDS
              /*            <== comment which, if first, becomes IDS
              @end          <== end of header (jump to Zdata read immediately)
              @skip n       <== skip next N lines of file
              @ <commands>  <== stack commands on process stack in LIFO order
              ZDATA         <== end header and start read of Zdata immediately
              COLS: | NCOL: <== set number of columns
              ROWS: | NROW: <== set number of rows
              COMMENT       <== becomes IDS of the surface
              XSCALE        <== X = xorigin + i*xscale (if XDATA not specified)
              YSCALE        <== Y = yorigin + i*yscale (if YDATA not specified)
              XORIGIN
              YORIGIN
              XDATA         <== Load NCOL values of X (column) header values
              YDATA         <== Load NROW values of Y (rows) header values
         Commands @end and @skip are case sensitive.
            
MODIFY:  Minor modification to handling of calls to an externally linked
         function from the expression evaluator when in complex mode.  See
         new documentation in "SimpleFnc" of the USER code directory.

MODIFY:  DEFINE of a name already existing as a DECLARE will now print an
         error message and change the class of the variable to FUNCTION
         (from STRING).

ENHANCE: Really a fix, but some may view as an enhancement.  Added translation
         of %var% to expression in &encode usage.  Following works as expected
         (after you think about it a bit).
               declare var = "ids"
               echo &encode "The value of %s is: %s" var %var%

ENHANCE: String manipulations are now fairly strongly supported in the 
         expression evaluator.  It has not permeated all elements of the
         code, but such things as:
           abbrev        center         char           concat         copies
           d2x           ichar          index          left           length
           LexEqual      lowcase        lowercase      pos            reverse
           right         strcat         strcmp         strcspn        stricmp
           strip         strlen         strncmp        strnicmp       strnlen
           strspn        substr         subword        translate      upcase
           uppercase     verify         word           wordpos        words
           x2d
         At least, now you get:
            GENPLOT: eval reverse(subword("This is a long line", 1, 3))
            Value: "gnol a si" = reverse(subword("This is a long line", 1, 3))

         Eval tries to ascertain the nature of the expression before printing.
         If the first reference is to a string variable, function returning
         string, or a string constant (any expression in double "), then it
         will evaluate it as a string.  If the first item appears to be numeric,
         then it will invoke the numerical interpreter mode.  

         Note that sloppy use of DEFINE and DECLARE will not cause problems.
         DECLARE always creates a STRING variable and DEFINE always creates
         a FUNCTION variable.  STRING variables are interpreted as functions
         if they are encountered in the middle of an expression.  But they
         are intrinsically strings and may switch modes on you when not
         expected.
              GENPLOT: declare value 7+8
              GENPLOT: eval value
              Value: "7+8" = value
              GENPLOT: eval 0+value
              Value:              15 = 0+value
              GENPLOT:
              GENPLOT: define total 7+8
              GENPLOT: eval total
              Value:              15 = total
              GENPLOT:
         
         GENPLOT: declare str = copies("hi", 4) eval str
         Value: "hihihihi" = str
         GENPLOT: declare str = "copies(""hi"", 4)" eval str
         Value: "copies("hi", 4)" = str
         GENPLOT:

ENHANCE: At behest of Mr. Smith, added some of the REXX string functions
         as well, along with their UGLY optional argument formats.  He has
         promised to document them properly.  Currently present functions
         return only numeric values, but the string returns are coming.
             abbrev(str1, str2, minlen)
             string(length)
             x2d(string)
             words(string)
             index(str, char [,start] )
             pos(str, char [,start] )
             wordpos(word, str [,start] )
             verify(str, valid_list [,start [,reverse] ] )

ENHANCE: Added some string comparison functions to the expression evaluator.
         Strings are still limited, but provides at least compare capability.
              strcmp(str1, str2)       /* String compare 
              stricmp(str1, str2)      /* String compare (case insensitive)
              strncmp(str1, str2, n)   /* Compare only first n chars
              strnicmp(str1, str2, n)  /* Compare only first n chars
              strlen(str)              /* Length of the string
              strnlen(str)             /* Length ignoring trailing blanks
              strcspn(str1, str2)      /* first element in str1 that matches
                                          any character in str2
              strspn(str1, str2)       /* first element in str1 that matches
                                          none of the chacters in str2
         These are identical to their C function calls.  Note that is means
         THEY RETURN 0 WHEN TRUE, and NON-ZERO when FALSE.  Direct tests
         will not work!

            BOOL LexEqual(char *string, char match, int minlen) 

         Also defined lexical equality for more common use.  This is the 
         same as the primary matching function in Genplot.   String is the
         input string, match is the expected pattern, and minlen is the
         minimum acceptable abbreviation length.  The strings must match
         for all given characters, although either string may have additional
         chars beyond the other.  This function is extremely useful for
         writing command case expressions.

         The string arguments may be either string variables, or quoted
         strings.
              eval strnlen("Who said ""Where no man has gone before""? ")
              eval strlen("Who said ""Where no man has gone before""? ")

              echo off
              while (1) {
                 declare cmd = &getarg -prompt "What's up boss? " -default quit
           
                 if LexEqual(cmd, "quit", 1) {
                    printf "You are not allowed to quit at this time"
                 } elseif LexEqual(cmd, "help", 1) {
                    printf "No help is available"
                 } elseif (strcmp(cmd, "really really quit") == 0) {
                    printf "Well if you really really insist ..."
                    break
                 } else {
                    printf "No commands are implemented"
                 }
              }


ENHANCE: Added linkage to the normalized error correlation matrix in nlsfit
         (using curfit method only).  This is linked as a simple array that
         must be interpreted manually as an NxN matrix.  

         Added command "NLSFIT ERROR_matrix" which will print the error
         correlation matrix to the screen (formatted reasonably).

         The error matrix is the inverse of the estimated curvature matrix
         for the X^2 hypersurface, considered as a function of the N varying
         parameters.  The square root of the diagonal elements of this 
         matrix give the estimated sigma's for the fit parameters.

         GENPLOT reports not the error matrix itself, but the normalized
         error matrix; specifically each element (i,j) is normalized by
         the product of sigma_i and sigma_j.  The elements the represent
         error correlation as relative in the reported sigmas.  All
         elements of the normalized error matrix must have absolute value
         less than 1.0, and the diagonal terms will be exactly 1.0.

         The values of the matrix may be accessed internally via the
         linked array CORRELATE$ which is a linear N^2 array.  Or the
         command NLSFIT ERROR will print out the matrix as in the 
         example below:

           NLSFIT: error_matrix
             Error correlation matrix (CORRELATE$)
               1.000   -0.001   -0.495    0.057   -0.100    0.106
              -0.001    1.000   -0.036    0.026   -0.003   -0.008
              -0.495   -0.036    1.000    0.165   -0.305    0.320
               0.057    0.026    0.165    1.000   -0.883    0.789
              -0.100   -0.003   -0.305   -0.883    1.000   -0.980
               0.106   -0.008    0.320    0.789   -0.980    1.000
           NLSFIT:
         
         Precise interpretion of the error matrix can be difficult.
         However, qualitative interpretation is fairly simple.  If
         an off-diagonal element is large, then errors in the two
         parameters are strongly correlated; changes in one parameter
         may be "compensated" for by changes in the second.

         The above results are from the fit of a peak using a gaussion
         with a quadratic background.  The parameters were the amplitude,
         center, FWHM of the peak and the offset, slope and quadratic
         components of the background.  From the values in the first
         column, we see that the amplitude and the FWHM errors are
         negatively correlated to a significant degree -- an increase
         in the amplitude can be compensated by a decrease in FWHM. 
         Quantitatively, increasing the amplitude by sigma and
         decreasing the FWHM by 0.495*sigma will result in a curve
         only slightly modified from the best fit.  

         Looking at other terms, the center is essentially
         uncorrelated with any of the other parameter.  The offset
         of the background, on the other hand, is correlated 
         significantly with both the slope and quadratic terms in
         the parabolic fit.

         In brief, the matrix allows one to construct the hypersurfaces
         of constant increase in X^2 from the minimum fit value, and
         thus the directions of the hyper-ellipsiods.  For more detailed
         interpretation of the inverse curvature matrix as an error
         correlation estimator, consult a serious data analysis textbook.

ENHANCE: Added ability to set the angle in annotate via coordinates (for
         macros or to follow a given line.  If a "/" is given to the angle
         command, the cursor or entry option is given.
           annote angle / /                   -> brings up cursor for 2 points
           annote angle / 1.0 1.0 2.5 3.7     -> sets angle for start to end

ENHANCE: CCIRCLE command in ANNOTE to draw a simple circle based on X,Y
         center position and a radius (in inches).
               annotate ccircle <x,y center> <radius>

MODIFY:  CREATE now permits npt to be 1.  The range is ignored and the single
         point will be created at the minimum value of any specified range.
         (Previously, "create -points 1" failed with an error message).

MODIFY:  The postscript driver now supports filling the color of the internal
         axes area, and page color filling.  Set pen color 0 (via COLORMAP)
         to the color desired inside the axes area (default is white).  Use
         PAGECOLOR to set the color of the page.  With PageColor, the bounding
         box is set to 0.2" beyond the drawing (a border of the page color).

         When directly outputting the page, the page fill is set to 26 inches
         wide and high to try to get the entire page.

         This is a tentative implementation -- let me know if there are 
         problems with conversion in word processing documents, etc.


ADD:     Color names now include page and background to represent these colors.
              Background -> 0  -- color of the inside of axes region, used to
                                  erase curves
              Page       -> 0 or value set by "pagecolor" command.
         These values are not terribly useful except on PM and postscript
         drivers which properly interpret the page and axes filling commands.

ADD:     Function depth is now limited to 200.  This should not affect
         real expressions but will avoid crashes due to circular mistakes.
            eval f(sin(x*2)) 
         has a recursion depth of 3 (x*2, sin, f).  Getting 200 deep ought to
         be very difficult unless you make a mistake.
            define f(x) = f(x)+1
            eval f(x)
         or
            define ival = 1
            let ival = ival+1
            eval ival
         In the latter case, because ival is DEFINED and becomes a string, the let
         is also a string define.  When eval comes along, it substitutes for ival
         -- but the substitution has ival in it, so it continues to expand
         indefinitely.
            eval ival --> ival+1 --> (ival+1)+1 --> ((ival+1)+1)+1 ....
         The function depth will avoid a stack overflow.  If this number turns
         out to be too low for anyone's work, easy to increase the threshold.
            
ADD:     After multiple requests by RdR, now have ability to change the file 
         used to define devices (devices.dat).  The syntax is:
               DEVice -Filename <PATH>
         This replaces the internal "devices.dat" name with the given name or
         path.  The file will be searched for in the current directory and along
         the normal Genplot system file search path (dynamic load path).

         Usage: DEVice [ dev_name | -option ]
          
         Options are identified by a leading hyphen and may or may not accept arguments
               -?               - quick help
               -OPTION <token>  - token passed to driver as driver specific command
               -TEXT            - request driver to display text window
               -GRAPH           - request driver to display graph window
               -NOGRAPH         - request driver to turn off graph window
               -NOTEXT          - request driver to turn off text window
               -CHECK           - if a device has been initialized, do nothing.
                                  Otherwise do a "device auto"
               -FILENAME <file> - change the internal name of devices.dat
               -QUERY           - prints out current device and device.dat filename
         For the device name, the name "auto" causes the GTERM variable to be
         scanned and used -- or uses a system default device.

         The modification to the devices.dat filename is sticky -- ie. it will
         not be undone by a GENPLOT RESET command.


MODIFY:  The ASCII file read now automatically expands the line buffer as
         needed to accomodate arbitrary long lines (and arbitrary numbers of
         columns per line).  This DOES NOT apply to reads within macros (using
         the << structure), but only when reading a real file.  It has been 
         tested to lines of 750,000 characters with 60,000 columns of data but
         should have no intrinsic limits beyond memory constraints.  Say thanks
         to Pat for writing such incredibly long data files ...

         A nessage us printed for each 1024 character expansion of the line 
         length just so you know it is happening.

         Technical: Applies to ASCII XY, XYZ, or list directed reads scanning
                    directly opened file.  All other modes remain limited to
                    1024.

MODIFY:  The following keys are trapped by XGENPLOT and XRUMP to scroll the
         text window:
               <PAGEUP>            - scroll by one window size
               <PAGEDOWN>          - scroll by one window size
         <CTRL><PAGEUP>            - scroll to top of display buffer
         <CTRL><PAGEDOWN>          - return to bottom of display buffer
         <CTRL>V                   - scroll down one page (emacs)
         <CTRL><DOWN>              - scroll by one line
         <CTRL><UP>                - scroll by one line

         These changes apply only to the WinNT version.  For OS/2, only the
         <PAGEUP> and <PAGEDOWN> are implemented.  Also, while <ALT><V> would
         be nice, all <ALT> characters are trapped by the window interface as
         menu commands.  <CTRL>V is there for those of us used to emacs.

ADD:     Filename operations INTERNAL to GENPLOT (most) will now ignore either
         single or double quoted filenames.  In addition, the UNIX form of
         "~/filename" will be translated to %HOMEDRIVE%%HOMEPATH%/filename"
         automatically.  Can't look up user information, so any form of
         "~user/filename" is similarly translated (ie. all usernames resolve
         to the current user).  Someday this may change.

MODIFY:  Changed the filename completion algorithm to better handle filenames
         with embedded blanks (can Microsoft do anything more annoying now?
         First we have the / versus \ fiasco, and now allowing spaces in 
         filenames -- talk about complicating a parsing algorithm!).

         CHANGE QUICKLY: The " is now only added only for filenames containing
         blanks.  Otherwise, just normal processing.

SLIGHT:  Both "foreach" and "loop" now support the -? or -help options to
         get quick help.

MODIFY:  The loop command now takes optional ranges ....

         loop %z=start,end[,increment]

      Usage: loop <count> [do | times]     <block statements>
             loop %x=<start>,<end>[,<inc>] <block statements>
    
         In the second form, the start, end and increment values must all be
         together in a single extended token -- ie. no spaces are allowed.
         Expresssions are allowed, but will be converted to nearest integer
         values.  %x will be substituted with a file extension format (ie.
         normally 007) where x may be any character.
    
         %c and %i are always substituted with the count and index starting
         from 001 and 000 respectively.

         Example: loop %a=91,113 { read basefile.%a plot hcopy dev pogo }

         
MODIFY:  The IDENTIFY commands now have numerous additional options to
         override almost everything.  Because of the way that parsing is
         handled, these options are only available to the IDS and IDENTIFY
         commands and not as part of the -IDENTIFY option to PLOT and OVERLAY
         (would conflict with the options to those commands anyway).

            IDENTIFY [text] [options]
           
               -PLACE { / | <x> <y> }    - Starting legend position in inches (cursor)
               -SKIP                     - Skip one space before the legend entry
               -CLOSE                    - Close the legend box after this entry
               -L&S                      - Do both lines and symbols on legend
               -SYMbol <sym>             - Override and use this symbol type
               -LType  <lt>              - Override and use this linetype
               -PEN    <color>           - Override and use this color
               -LWidth <linewidth>       - Override and use this linewidth
               -NOMARK -NOSYM -NOLINE    - Draw text only, or turn off each
               -RESET                    - Reset to top of legend on new plot (internal)

         In addition, if doing -L&S, the line segment example is drawn from 
         the symbol out in the two directions so as to not go through the
         symbol.  It is likely that $IDLSIZ or $IDVSIZ should be changed if
         doing extensive lines and symbol markers (increase $IDLSIZ or
         decrease $IDVSIZ).

         Note that -NOMARK disables both lines and symbols, while -NOSYM
         and -NOLINE only disable that specific marker.  If you draw with
         symbols and indicate -NOLINE, it will have no effect.

         Finally, multiple line LEGEND entries are now permitted with the
         \n separater in the text.


MODIFY:  The foreach (and the for) commands will now accept alternates to the
         %f variable.
                foreach [%x [in]] (list,...) [do] <executable block>
         The x in %x can be any character, defaulting to f if not specified.

         foreach %a (a b c d e) { foreach %b (f g h) echo %a%b }    (minimal)
         foreach %a (a b c d e) do { foreach %b in (f g h) do echo %a%b }
         foreach %a in (a b c d e) do { foreach %b in (f g h) do echo %a%b }

         The "for" command continues to be non-rentrant, though I've seen no
         problem with foreach and thus soon expect to have it translate to
         foreach.

CHANGE:  The "default" device when none has been selected in OS/2 and NT is
         now "pm".  I'm getting tired of typing that.  GTERM will still
         override if it is set as an environment variable.

COMMENT: OS/2 is slowly leaking out of the code in name.  Made a check on
         the trade rags to see how often it has been mentioned in the last
         8 months and came up with only 5 valid comments on OS/2 WARP!  When
         will IBM ever learn?

NEW!!!: Added serial number verification -- had to at sometime since
        some users actually get charged for the program :-(.  This is how
        it will work -- 

           (1) Purchased licenses will get a serial number that permanently
               enables any update revision of the code.  This serial number
               is printed in the "About ..." help button and the "VERSION" 
               command.  Genplot only serial numbers will not run RUMP, 
               although RUMP serial numbers can also run Genplot.
               
           (2) Beta test, or test and evaluation users, can download or
               install the code, and with the right password, run until 
               some expiration date (60-90 days if I stay on top of it).
               A quick message giving the expiration date flashes by during
               bootup (2 second).

               After the expiration date, you must return and get a new
               copy and possibly evaluation serial number.  The old version
               will continue to work, but it will put up a message and
               force you to endure a 60 second timeout.

           (3) All users may regularly update their code by downloading the
               latest installation program from the WEB.  These updates will
               normally include the entire package. (The documentation may be
               split into a separate file if it gets too big.) Unfortunately, 
               licensed users will have to re-enter the serial number during
               each upgrade install.

           (4) This license verification currently only applies to Windows
               NT/95 versions.  UNIX users have the source code and can 
               bypass any software check anyway.  OS/2 users will ultimately 
               be forced to NT and, until that time, will be granted a free
               license.

        This mechanism seems to ensure that everyone can continue to get
        immediate updates and fixes, while providing some mechanism to 
        control the distribution of the program.


MODIFY: Added in NTDRIV (PM for NT only!!!) the ability to request metafile
        or clipboard output from a macro using the IOCTL capability.  Format
        is
            dev -option print
            dev -option -print                    w/ Powerpoint kludge
            dev -option clipboard"
            dev -option -clipboard                w/ Powerpoint kludge
            dev -option "metafile:<filename>"
            dev -option "-metafile:<filename>"    w/ Powerpoint kludge
        Notes:
           (1) The form "metafile <filename>" must be enclosed in quotes
               since it must be sent as a single string.  The filename is
               optional, but if not given, will bring up the dialog box.
               The filename must be fully formed -- no extension will be
               added to the specified string.
           (2) The text for all of the forms CANNOT be abbreviated -- ie.
               it must be all 8 characters of the word metafile.  
           (3) Errors are silently ignored -- the "dev -option" mechanism
               is specifically device dependent and option strings on
               one device may have no meaning on another.
           (4) For metafile with no specified filename, focus is shifted to
               the graph screen so the dialog box is visible.
           (5) Print always go to the default printer.

        Examples:
           dev -option "metafile e:\thesis\figs\chap1.fig.1.emf"

        GRPRINT command now behaves the same way -- essentially
        "dev -option print" is synonymous with the "grprint" command.

        Someday there may be a "EMF" device specifically as a subset of
        the PM screen driver (bastardize the final PMDRIV code).  However,
        at the moment, things change too quickly to try to keep two copies
        of very complex code maintained.

MODIFY: Added ability to write metafile from PMSCREEN with the embedded
        Powerpoint bug workaround.  Exists as another icon on the toolbar
        and as a menu item.

ENHANCE: Modified the handling of the ok; (system) commands for OS/2 and
         NT.  If the command line is now terminated with an & character,
         it is interpreted as a request for a detached session (independent
         of the running GENPLOT session.  This is equivalent to prefixing
         the command with "start".  So "ok; start eps" is identical to
         "ok; eps &".  Also, a blank request to ok to start a shell is
         interpreted as starting a new window if running in GUI mode
         (XGenplot or XRUMP).  In window mode (Genplot or RUMP), this is
         is still a inferior command process running in the current window.

         Finally, the EDITOR command (using the environment variable EDITOR)
         is started as a detached process in both windowed and GUI modes 
         under NT and OS/2.  Editing within the GENPLOT text window is no
         longer really productive.  

         For NT, the editor now also defaults to NOTEPAD which is usually
         on the path.  WORDPAD would be a better alternative, but Microsoft
         chose not to put it in a location on the path so I cannot be
         guarenteed to find it.  To use WORDPAD, you must set the EDITOR
         environment variable to the ENTIRE path to the executable.

ENHANCE: Added "SETLOCAL/ENDLOCAL" local variable control.  The command 
         setlocal increments an internal variable and all new variables
         added at this level are deallocated automatically when the 
         corresponding "ENDLOCAL" is executed.  These may be nested.  The
         command "GLOBALize" takes a list of variables and converts them,
         if necessary, from local to global.  The "LISTV" command lists 
         the level of each variable just before the name; 0 is global.

         This feature must be used with care.  It is primarily meant for
         very complex macros which are well debugged.  Existing variables
         level will not be modified.  But archiving to an existing curve
         name will delete the underlying curve and replace it with one
         at the local level -- which will be released on ENDLOCAL.

         Future features: (1) SETV will selectively create a new variable of 
                              the same name if one already exists at a more
                              global level.
                          (2) GLOBALIZE at this point will 

ENHANCE: Added ability to independently set the 4 margins around the page,
         letting the top and bottom differ, as with left and right.  The
         values are set by modification of the "MARGINS" command to correspond
         with the RANGE commands.
            MARGINS { [BOTTOM LEFT TOP RIGHT <value> }    or
            MARGINS <x> <y>
         In the latter case, the X and Y values set the left and right equal
         and the top and bottom equal.  This retains compatibility with the
         previous usage of the MARGIN command.
         The PAGE LAYOUT dialog box in NT/95 has the additions.  It is not
         present in OS/2 -- the current dialog box value modifies the margins
         in the compatability mode.
          
DOCUMENT:  The TRANSFORM DY/DX is a 3 point curvature determining derivative.
           If the data is not strictly sorted, there is a potential for 
           impossible calculations (3 local points with the same X value).
           Not wanting to prejudice the results, the DY/DX routine simply
           returns 0 for these invalid conditions.  To avoid this with good
           experimental data, SORT -STRICT or addition of random noise into
           X is highly recommended.

FIX/MODIFY: Behavior of the SORT command is fixe and options documented:
            SORT [options]
                -normal        Not reversed order
                -reverse       Reverse whatever else is said
                -nosort        Don't actually do any sorting
                -random        Randomize instead of sort
                -strict        Delete adjacent identical points - keep first
                  -delete      Default behavior (synonymous with -strict alone)
                  -average     Strict sort with neighboring points averaged
                  -avg         Synonymous with -average
                -xonly         Modify the X array only (implicit sort on X)
                -yonly         Modify the Y array only (implicit sort on Y)
                -zonly         Modify the Z array only (implicit sort on Z)
                -x             Sort using the X array
                -y             Sort using the Y array
                -z             Sort using the Z array
            These should behave closer to expected for multiple options.

            There is some interaction between the options.  In particular, the
            array used for testing will be sorted (if SORT requested).  Setting
            -X implicitly sets request to modify the X array.  A subsequent
            -yonly will change to sort to based on Y only.  On a 2D array,
                sort -yonly -x 
            is synonymous with simple sort since -yonly forces the Y array to
            be modifiable, and the -X forces the X array modifiable with sort
            testing on X.

            The one ambiguous mode is "sort -z -strict".  In this case, the Z
            and X arrays are taken as the "test" values, with Y the dependent
            variable.  Thus -average will average the Y values if both X and
            Z are equal.  The testing assumes the Z array is in order.

CHANGE: Modified the behavior of $csmin (minimum size of tick mark labels on
        the axes).  Behavior now dependent on the sign of $csmin:
           If positive: True minimum size.  If the "determined" size is less 
                        than this value, it is automatically bumped up in size
                        regardless of the physical validity.  
           If negative: Minimum cutoff size.  If determined size is less than
                        the absolute value, labeling is disabled.
        Note that the old behavior is as if negative - and the new default
        value is -0.07 inches.

SCREWED!! Due to critical bugs in the 5.0 compiler, all optimization has
          been disabled in both GENPLOT and RUMP.  I don't know which 
          routines are subject to error, but it is pervasive enought that
          prudence dictates disabling the speed optimization.  This may be
          as much as a 50% speed penalty in worse case (ie. FFT -POWER).
          If/when there is a patch released, I may try optimization again.
          Reminded yet again why I preferred IBM to Microsoft.  What a pain
          in the proverbial posterior tissue.

---------------------------------------------------------------------------
Now using MSC version 5.0 (really 11.00 on itself).  Minor changes required
to system.c for environment support.  Little else.  Should be okay.
---------------------------------------------------------------------------

Added:  Added -SILENT option to the cursor command.  Mostly for macros that
        give their own prompts and handle results through the set variables.

Modify: Added ability to change the repeat (vecsize) for the ID via the
        SGRAPH command on $IDVSIZ.  Default is 0.003.  But did increase the
        default line-length to 0.65 (from 0.40) to get most of the repeat
        cycle on screen.

Unmodify: The pattern size (repeat loop for - - - types) is not 0.001 instead
        of 0.003 in the IDS.  Makes it easier to see the line type.

Modify: New color on the PMSCREEN Color Map (OS/2) called NONE.  This is
        primarily for disabling the AXES background fill.

Modify: Potentially changes in the handling of options to plot.  They are
        now nominally coherent for all 2D operations (not 3D yet).

   Plot and Overlay modes:
   
   The overlay command is one of the most complex with various modes.  It is
   easiest to think about the options in two terms -- first options that
   specify the plot type (types) and second options that modify the character 
   of the plot (modifiers).  Not all modifiers apply to all types, but in
   general most do.
   
   ===================
   Types:
     -AZIZ        draws connected lines between adjacent points (in 
                  honor of Mike Aziz at Harvard University.  Ancient 
                  way to get error bars.  This mode ignores all lines
                  and error bars.
   
     -PT_LABEL    draws the ordinal number of each point -- useful 
                  when you are trying to follow trends in sequences.  This
                  mode ignores all lines and error bars.  Obviously the 
                  IDS will have a problem -- choose to show as lines.
   
     -HISTOGRAM   lines connect points as constant Y segments, beginning
                  halfway between adjacent X values.  By default, the
                  histogram is line only (current linetype).  Symbols and
                  error bars may be specified.  There is no exclusion of
                  histogram line around points.
   
     -BARGRAPH    Histogram as above plus lines extending to the X axis.
                  By default, the bargraph is line only (current linetype). 
                  Symbols and error bars may be specified.  There is no
                  exclusion of histogram line around points.
   
     -STICK       Vertical line segments from X-axis to the points. By 
                  default, this is a line only (current linetype). Symbols
                  and error bars may be specified.  There is no exclusion of
                  histogram line around points.
   
     <default>    Line or symbol plot
   
   ==================
   Modifiers:
     -SYMBOL <n>        Override the symbol type.  Either an ordinal or
                        name (ie. FilledCircle) can be specified
   
     -LTYPE  <n>        Override the linetype number.
     -LINETYPE <n>
   
     -PEN <n>           Override the pen color.  Either an ordinal or
     -COLOR <n>         name (Red) may be specified.
   
     -SYMSIZE <r>       Override the symbol size.
     -SYMBOLSIZE <r>
   
     -LW <r>            Override the linewidth
     -LINEWIDTH <r>
   
     -L&S               Where relevent, do both lines and symbols
   
     -IDS               Show IDS for this curve
     -NOIDS             Don't show IDS for this curve (even in AUTOIDS on)
   
     -IDENTIFY <text>   Draw IDS for this curve with alternate text
   
     -ERRX      <expr>  Error bar extent (symetric)
     -ERRXMinus <expr>  Error bar in X on minus side
     -ERRXLow   <expr>
     -ERRXPlus  <expr>  Error bar in X on plus side
     -ERRXHigh  <expr>
   
     -ERRY      <expr>  Error bar extent (symetric)
     -ERRYMinus <expr>  Error bar in Y on minus side
     -ERRYLow   <expr>
     -ERRYPlus  <expr>  Error bar in Y on plus side
     -ERRYHigh  <expr>
   
     -XSCALE            Draw as if the X axis had been drawn with autoscale
     -YSCALE            Draw as if the X axis had been drawn with autoscale
     -XYSCALE           Draw as if the X/Y axes had been drawn with autoscale
     -RESCALE           Synonymous with -XYSCALE
   
     -XRANGE <l> <h>    Plot as if the X scales are as specified
     -YRANGE <l> <h>    Plot as if the Y scales are as specified
     -ZRANGE <l> <h>    Plot as if the Z scales are as specified
     -XYRANGE ...       Plot as if the X & Y scales are as specified
     -XYZRANGE ...      Plot as if the X, Y & Z scales are as specified
   
     -PLX [Bottom|Top]  Override so plotted against specified scale
     -PLY [Left|Right]  Override so plotted against specified scale
     -BOTTOM            equivalent to -PLX BOTTOM
     -TOP               equivalent to -PLX TOP
     -LEFT              equivalent to -PLX LEFT
     -RIGHT             equivalent to -PLX RIGHT
   
     -SLOW              Try to slow down the drawing speed
     -SLOWER            Give 30 ms between each drawing point
     -SLOWN <ms>        Give specified # of ms between drawing points
   
     -EXCLUDE [-SELF | <curve> ]
                        Exclude line drawing over specified curve points
   
   Notes: (1) Specifying both -SYM and -LT will cause plot to do both lines
              and symbols (equivalent to -L&S).  -SYM or -LT alone will 
              override to give the desired effect.
          (2) Errorbars and symbols can be drawn with any plot type except
              AZIZ and PT_LABEL.
          (3) Exclude is primarily for normal lines or symbol plots.  Usage
              in other Plot Types has not been extensively tested.
          (4) HISTOGRAM, STICK and BARGRAPH implicitely select lines only
              (default = 1) unless -SYM is explicitely given.

Modify: The pattern size (repeat loop for - - - types) is not 0.001 instead
        of 0.003 in the IDS.  Makes it easier to see the line type.

Fix: XEQ -RETURN from within a memory block now returns out of the last
     macro call rather than just the block level.  Following used to fail:
        if (1) xeq -return

Enhance: Added EXEcute as alias for XEQ

Fix: I've given up and taken the route of fully deleting and recreating
     the device context on an erase in NT.  This should eliminate the
     ghosting characters remaining after erase on the screen.

Enhance: Added RANGE commands to NLSFIT so comparable to other fitting
         routines.  However, since NLSFIT is command driven, they are just
         a tad different.
            XRANGE | RANGE  - limit on X range values
            YRANGE          - or/also on Y range values
            ZRANGE          - or/also on Z range values (obviously 3D only)
            XYRANGE         - obvious I hope
            XYZRANGE        - and given above, even more obvious
            CURSOR_RANGE    - show a box cursor and select valid region         
         Note that the range can also be specified giving an error bar
         expression which has 0 entries for undesired points.

Fix: LSQFIT now uses same definition for option [-RANGE -XRANGE -YRANGE
     -XYRANGE] as other FIT commands.

Enhance: Added -XYRANGE to all FIT commands.  Now have
             -XRANGE (or -RANGE synonymously)
             -YRANGE                                                  NOT NEW
             -XYRANGE (equivalent to -XRANGE -YRANGE)                 NEW
             -CURSOR (puts up box cursor and selects allowed window)  NOT NEW
             -WEIGHT (may be ignored by some fit routines)
             -SIGMA  (may be ignored by some fit routines)
             -SILENT (may be ignored by some fit routines)
         fit <type> <type_opts> [-XRANGE <xmin> <xmax>] [-YRANGE <> <>]
                                [-CURSOR] [-WEIGHT] [-SIGMA] [-SILENT]
         on some.

Add:     New options to PLOT:
            -NOIDS             (Both RUMP and GENPLOT)
            -XRANGE <xlow> <xhigh>
            -YRANGE <ylow> <yhigh>
            -ZRANGE <ylow> <yhigh>
            -XYRANGE <xlow> <xhigh> <ylow> <yhigh>
            -XYZRANGE <xlow> <xhigh> <ylow> <yhigh> <ylow> <yhigh>
         The -RANGE force the appropriate range to be as specified on the
         given axis without making any permanent or otherwise change.  
         Affects ONLY the plotting of the data but not the AXES etc.

Add:     Added code for proper handling of DOUBLE and INT arrays from
         within the expression evaluator.  No way to create them from
         inside GENPLOT, but they can be linked from external user routines.
           BOOL GVLinkArray       (char *name, int flags, REAL x[],    INT maxsize, INT *size);
           BOOL GVLinkIntArray    (char *name, int flags, INT i[],     INT maxsize, INT *size);
           BOOL GVLinkDoubleArray (char *name, int flags, DOUBLE x[],  INT maxsize, INT *size);
           BOOL GVLinkComplexArray(char *name, int flags, COMPLEX z[], INT maxsize, INT *size);
         In user usage, the simplest form is to set "flags=0" and "size=NULL".
         Setting size to NULL causes the size to always be maxsize.  Setting
         it to a static variable address would allow you to have the dimension
         of the array change dynamically.
          
Document: sprintf <name> <format> <vars ...>
         Creates (if necessary) or sets a string variable to a formatted
         string from internal variables.  Similar to 
            declare <name> &encode <format> <vars ...>

Add:     For binary writes, now can embed commands also via write.  Format
             write <filename> -binary -commands
                <cmd1>
                <cmd2>
                <cmd3>
                @end
         -command can be replaced with -text or -comments (plural only) to
         embed only comments, not commands.  The -COMMENT <ids> option gives
         the identifier line on the string.

New:     -IDS is synonymous with -COMMENT in a WRITE statement (preferable!)

Add:     In binary data format (GENPLOT), comments in the header beginning
         "@ " are considered to be command requests and will be queued for
         executation in an RPN order following the read (ie. last read will
         be executed first).  This is now consistent with the ASCII read
         which has the same construct.

Add:     Because NT is so brain dead, had to modify the I/O channel to
         support LPR output to TCP/IP connected hosts.  New channel:
             LPR*"printer"
             LPR*"-Sserver -Pprinter"
         This I/O channel creates a temporary file for the plot and then
         spools it via the LPR command.  Unfortunately, the NT version of lpr
         cannot take the file from standard input (no piping) so this is the
         only way.  Under UNIX and OS/2, this is mapped to be equivalent to
         PIPE*"lpr -...."

Modify:  NT Genplot windows size can now be specified -- ie. a buffer space
         for scroll back information.  This information is in a new .INI
         file Genplot\NT\genplot_.ini under the section [CONSOLE] with the
         keys ROWS and COLUMNS.  If you choose to modify these sections,
            (1) copy genplot_.ini to the Local subdirectory and modify it
                there, recognizing that new sections will be added in the
                near future and you will have to manually add these sections
                to your new copy
            (2) modify the original genplot_.ini, recognizing that at each
                new update, you will have to remake these changes.
         Sorry for the pain - part of the curse of development testing.

Modify:  To be consistent, FIT and TRANSFORM now default to ABORT instead of
         LIST.  Also, ? is synonymous for LIST, and FIT/TRANSFORM return to
         GENPLOT prompt after printing the list (used to recycle asking for
         the fit/transform type).

Modify:  The internal MORE routine (used by TYPE) now accepts <ESC>, ^Z, ^G
         as synonyms for Q (quit).

Enhance: Because of the popularity of "implicit" let commands, the X command
         now checks for an = sign.  If present, it is an implicit set of
         the X variable.  Otherwise it is a call to XEQ as before.
            CREATE Y = SIN(X) -range -pi pi
            X = COS(X)
            PLOT

Enhance: In GENPLOT, a filename of the form "zipfile::file" will be properly
         read from the zipfile.  The file must be fully specified (ie. with
         any trailing .dat) and the zip file must exist.  Few if any errors
         are reported.  Internally, it is expanded as:
             read archive::file1.dat
             read "| unzip -p archive file1.dat"
         The IDS will be archive::file1.dat unless reset by the read.  Like
         other PIPE reads, the input is assumed to be ASCII unless the read
         specifically specifies binary.
             read archive::file1.dat -binary

         The same file specification applies to XEQ macros and RUMP
         files (I hope).  The full macro name (including the .mac
         extension) must be specified.

         However, for RUMP, a .RBS extension will be appended if none exists
         on the given name.  This means files without an extension will be
         difficult to read from an archive.  Depending on the implementation
         of unzip, "read archive::file." may work to get "file" without an
         extension.  But it doesn't under OS/2.  Simple solution -- if you
         want to use the archive, accept the default extensions!  Direct read
         specified by pipe does still work 'read "| unzip -p archive file".

Enhance: Show command has been expanded as per request by WLB
           show [-xrange | -yrange | -zrange | -irange] <low> <high>
           show  -for <condition>
                 -xrange assumed if no option specified

Document: Identify has the following options:
      -place    --> sets location
      -skip     --> leaves a blank position in the legend
      -l&s      --> show both lines and symbols
      -nomark   --> don't draw a marker on the legend
      -nosymbol --> same as -nomark
      -noline   --> same as -nomark
      -close    --> draw a box around the legend
      -reset    --> reset for drawing from the beginning of the legend

Enhancement: Added a -SINGLE option to cursor commands.  This causes the
             cursor to return immediately on the next key press (left or
             right mouse, or any key -- including space).  Useful when you
             want to get a point in a macro from users who don't read
             the screen prompts.  (Press RIGHT MOUSE BUTTON??)

Enhancement: Added a "DOUBLE" type variable in addition to "REAL".  It is
   really double precision, but its use is slightly restricted.  For example,
   it cannot be used as a variable in "NLSFIT".  But it can be used in 
   any normal expressions or functions.  Primarily added to allow constants
   and temporary expressions which lie outside the 1E37 range of normal
   floats.

      alloc double <varname>
      setvar double <var>

   There may ultimately be a "PRECISION DOUBLE" capability for curves as
   well as variables, but at the moment that is painfully difficult and
   has not been done.

   Range is 1E+/-308, about 15 decimal digits


Modification: FIX_GRID majorly rewritten:
    Fix_Grid is used to impose a specified X grid on a given curve, 
    interpolating as necessary.  Either linear or spline interpolation can
    be used.  The X-grid can be specified as an equally spaced grid of a
    specified number of points between two limits, or can be specified as
    the abscissa of another curve (or an array).  Finally, outside the range
    of the data, either CONSTANT or ZERO extension may be specified.

    Usage: FIX_GRID <-options>

    Options: -RANGE <xlow> <xhigh>   Specify the range of the new abscissa
               -FROM <xlow>            grid.  FROM and TO overwrite the values
               -TO <xhigh>             of RANGE.  Initial are min/max of curve.
             -POINTS <npt>           Number of points in the grid.  Initially
                                       set to number in the curve.
             -MATCH <curve | array>  Causes the X-grid to match the specified
                                       curve.x or array values.  This option
                                       takes precedence over -RANGE and -POINTS
             -SPLINE                 Cubic spline interpolation used between
                                       given points of the curve.  Default is
                                       linear interpolation.
             -ZERO | -CONSTANT       Either constant extend the function, or
                                       set to zero outside valid range.

    Notes: An implicit SORT -STRICT -AVERAGE is done prior to grid fix.
           The original curve is replaced by the new gridded version.


Modification: Added ! as unary NOT operator.  !1?7:2 returns 2

Modification: OS/2 XGenplot and XRump now interpret a Button 3 click as
     as "Paste" command.  Same as Button 2 double click.

Modification: CLIPSYMBOL NO
     Previously, clipsymbol applied only to the symbol.  It would not be
     clipped even if it extended outside the plot box (but any symbol whose
     value was outside the box would be eliminated).  This did not apply to
     the error bars on symbols.  Now, it applies both to the symbol and
     to the error bars.  With CLIPSYMBOL NO, any point whose value is within
     the plot ranges will be drawn fully, as will its error bars.

Enhance: Need to document PageSplit command.
          PageSplit Set <# rows> <# cols>      
               Initializes multiplot page with given # down page and across
               Orientation set by current LANDSCAPE or PORTRAIT orientation.
                 "Orient Portrait  PageSplit Set 5 3"    Nice 15 plot page
                 "Orient Landscape PageSplit Set 3 4"    Other nice way
          PageSplit Select <row> <col>
               Make next plot start at specified position.  Both are taken
               mod the actual number of rows and columns
          PageSplit Cancel
               Returns to single plot per page format
          Erase (implicit or explicit)
               Advance to next plot position, going left to right and then
               top to bottom (row major format).  [Western Culture dictates].
          NewPage 
               In multipage mode, starts a new page.  In single page mode,
               is equivalent to the old erase.

Enhance: If a command is not recognized, the default is to attempt the
         operation as a "let", "setv", "define", "declare" or "eval"
         depending on the nature of the command line before returning an
         unrecognized command error.
            Subsequent token exists and is an "=" sign (space separated):
              Variable exists:
                 Set value of variable as per the expression following
              Variable does not exist:
                 Variable contains an parenthesis, treat as "define"
                 Expression quoted (explicit or implicit), treat as "declare"
                 Otherwise "setv" as REAL or COMPLEX based on MATHMODE
            Token contains a mathematical operator (+-*/ etc.)
                 Treat as "eval"
         In all cases where the variable is implicitely allocated, a message
         identifying the variable and the type that was assumed is printed.

         All of the &getarg processor commands return implicitely quoted 
         strings and hence will "declare" a string variable.  This can be
         a problem if you need the value to be "REAL" for use in operations
         like NLSFIT.  The following demonstrates the problem.
            f(x) = a*x
            a = &getarg -prompt "Initial guess of slope" -default 1
            nlsfit
              equ f
              vary a /    <==== FAILSE since a is a string, not a variable
         Math expressions are often okay either way since strings are
         evaluated as if they were functions of no arguments.  But it is
         safer to set variable explicitely with "setvar"  -- for the 
         following eventual reason:
             declare a = 7+8              /* A string declaration
             declare b = 8+9
             eval a+b                     /* Resolved to be a string expression
                 result: 7+8+8+9
             eval 0+a+b                   /* Resolved to be a numeric expressoin
                 result: 32

Enhance: New file tests (function evaluator):
           isfile(pathname)   returns 1 if path is an existing file
           isdir(pathname)    returns 1 if path is an existing directory
           filedate(pathname) both return seconds since epoch, or -1 if
           filetime(pathname)   the file does not exist.
           filesize(pathname) size of file in bytes, or -1 if doesn't exist

Document: GENPLOT Configure
         ? | help                                - obvious
         status | parms | parameters             - prints current settings
         value                                   - binary flag set (debug only)
         confirmoverwrite | noconfirmoverwrite   - check before writing file
         binarywrite | nobinarywrite             - default to binary files
         asciiwrite  | noasciiwrite              - default to ascii files
         datapath | dpath                        - set data file search path
         dataexts                                - set data file extensions

Enhance: Added two new key commands
            <ctrl><left arrow>  - backward word
            <ctrl><right arrow> - forward word
         There are currently no equivalent UNIX key strokes, but they can be
         defined via the termkey file -- see "termkey" in bin distribution.
         The definition of a word is my choice -- any other logical one will
         be happily considered (ie. does an _ define a word ending or just
         a normal character).

Enhance: New loop control commands
             break    - breaks out of currently active loop, foreach or while
             continue - skips to next element of active loop, foreach or while
         These control signals will only operate on foreach/while/loop blocks.

         GENPLOT: loop 200 setv i1=%c if (%c>=20) { continue } setv i2=%c
         GENPLOT: printf "i1=%d i2=%d" i1,i2
         i1=200 i2=19
         GENPLOT:

         GENPLOT: loop 200 setv i1=%c if (%c>=20) { break } setv i2=%c
         GENPLOT: printf "i1=%d i2=%d" i1,i2
         i1=20 i2=19
         GENPLOT:

Change:  The "block" statements within a "loop", "while", or "foreach" command
         are now by default made non-echo.  If you need echoing to debug, add
         explicit "@echo on" to the block.  This change also apply to all
         "xeqs" inside the block.  "if" blocks inherit the echo-characteristic
         of the calling program.  Change necessary because of the large amount
         of screen output that may occur from even a simple loop command.

Enhance: Genplot will now automatically write .gz files if the extension is
         detected.  write "test.dat.gz" becomes "| gzip -9 > test.dat.gz"

Clarify: Label statements cannot be dynamic.  :xxx must be a static name.
         This is required since a GOTO within a block "loop" or "foreach"
         will only scan the block once before rewinding to the beginning
         of the file.  GOTO's within a block will be properly handled,
         but only when the name is static.  :count%c  is invalid
         Segment below is an example of valid internal and external goto.
             loop 10 {
                setv icnt = 0
                :start
                echo %c
                let icnt = icnt+1
                if (icnt==1) goto start
                if (%c > 8) goto end
             }
             bad_statement_1
             bad_statement_2
             :end
             echo We are finished.
         Also, on console input, multiple blank lines (about 5 or 6 of them)
         will terminate a GOTO$ search (just so you don't get too frustrated)

Enhance: Added block execution blocks plus "loop", "while" and if/elseif/else
         constructs.  These are alpha test and hopefully haven't destroyed
         any other components.
            if (expr) {
                list of commands
            } elseif (expr) {
                list of commands
            } else {
                list of commands
            }

            while (condition) { 
               expressions
            }

            loop (count) {
               epxressions
            }

         Expressions not using { } delineation will assume the remainder
         of the current command line.  As usual, line boundaries are not
         relevent and everything can be on a single line.
           if (8<9) { if (4<3) { echo y^2 } else { echo y/n } } else echo no
         If blocks may be nested to a maximum depth of 64 -- but this includes
         all types of nesting including macros calling macros and for %f loops.
         Still, it is big enough that I don't think it should be exceeded too
         soon.

         WARNING: Syntax is not checked -- and certain structures are allowed
         syntactically though they make little sense.
            if (x < 8) echo hi
            let x = x-1
            elseif (x<8) echo another hi
         elseif do not have to immediately follow an if, and multiple else/
         elseif's may follow a single if.  Only one block will ever be
         executed and the flag indicating that one has been done is not
         cleared except on an if.  I don't want to know what craziness
         someone is going to do with this "feature" behavior.
         
         WARNING: At the moment, behavior of block statements with the
         "for %f" processing will not work.  See below for improvement.
 
         WARNING: while (condition) is tested BEFORE executation of the block.
         If the condition is false to begin, then the block will never be
         executed.  If there is interest, I can make an until (condition)
         command.

Enhance: for %f and repeat commands have been replaced.  The old will remain
         until I can ascertain that the new ones work, but then will be
         obsoleted.  New are flexible enough to work as old.
            foreach (a b c ...) { expressions }
            loop <n> times { expressions}
         Also works
            foreach %f in (a b c) do { expressions }

Enhance: xeq and read can now fully utilize pipes.  Both will now check if
         there is a file with .gz extension and use "gzip -dc" to unpack
         the file during execution.
Enhance: Added pipe cabability to LEX file execution (xeq "| rr") which reads
         input from a pipe started with the given command.  Also added pipe to
         multi-line read in ANNOTE.  multiline x,y "| test" as well as 
         multi-line x,y "< test.fil".
Enhance: 3D plots to the POSTSCRIPT driver now uses true fonts instead of
         the hershey fonts.  This only applies to the postscript driver --
         even the PM driver does not have general transformation
         capability as far as I can determine.

Enhance: ASCII data file reads have been sped up by a factor of 3-4 at the
         cost of a slight slowdown (5%) for data files with expressions
         rather than simple numbers.  Most of the data reads are of pure
         numbers which can be interpreted very fast.

Added: New functions
             exists(var_name) 
             limit(value, low_limit, upper_limit)
       and modified function
             typeof(var_name).
       exists() returns 1 if the variable exists and 0 otherwise.  typeof
       used to give an error if the variable did not exist -- it now returns
       -1 to indicate no variable type.
             limit(x,xmin,xmax) = max(xmin,min(x,xmax))
       which I always end up coding wrong.

Modified: Increased size of almost all strings.  There are now four general
          lengths.  80 characters if nothing but simple text.  256 for most
          general work.  1024 for anything that might need to be long (text
          output, equations).  There are some still 32 characters but only
          where the input is expected to be a variable name.

Modified: let command now aborts on errors rather than trying to complete
          a list expression.  Necessary because now accepts the form 
          "let p() = 1,7,3,5" to fill an array.  If () is specified, it
          is recognized as a request to enter all of the elements.

Added: More explicit directions and include files for compiling extensions
       with Visual C from IBM.  Set genplot/os2/user directory for
       information.

Modified: Causes L&S to give proper legend in GENPLOT.

Modified: Added -RESET to the IDENTIFY options.  This resets the next 
          position for a IDS to the "starting" point for next object.
          Useful for moving the starting point and then beginning a new
          set of idents on the page.

Modified: On cursor -track, HOME and END now move to beginning and end of
          the data set.  Also modified so it will not go off screen.

Document: Make sure to add ncol and nrow as reserved names for GENPLOT.

Added:   New command "MATRIX" similar to TRANSFORM with multiple subcommands
            TRANSPOSE       <source surface>
               Transposes the matrix (x<->y)
            DUPLICATE       <source surface> <destination surface>
               Duplicates a surface
            ROW_EXTRACT     <source> <row/rows>
            X_SCAN          <source> <row/rows>
            COLUMN_EXTRACT  <source> <col/cols>
            Y_SCAN          <source> <col/cols>
               The specified row/cols are extracted and averaged.  Either
               a single column can be specified (ordinal 1-ncol), or
               multiple columns in a parenthentical list (1,7) meaning
               average the first 7 columns.  NOTE: First row is row 1, and
               averaging first seven rows is done by
                 matrix row_extract 1,7
               Result is always left in the main curve (default)
            TO_CURVE        <surface name>
               The SURFACE is expanded into the default curve
            TO_SURFACE      <surface name>
               The current 3D curve is converted into a SURFACE assuming
               that it is really on a mesh.  X,Y points must make sense as
               either a column or row major matrix orientation (ie. X same
               for nrow Y values, or Y same for ncol X values).  Will fail
               if no underlying matrix is found.

         All MATRIX commands requests will use as the <source>, if expected,
         a currently active surface requested after the MATRIX command.  So,
         the following commands are synonymous:
              MATRIX S1 TRANSPOSE
              MATRIX TRANSPOSE S1
              MATRIX -SURFACE S1 TRANSPOSE
         assuming S1 is a valid matrix.

         Additional commands such as SQUISH, CULL, SMOOTH, etc. can be
         implemented as necessary in this formalism, upon request.

Added:   Can do a "status" on a matrix, getting something useful back.

Added:   Also, now can do an "eval" on surface or curves, with expected
         results.  The printout can be aborted with ^C.

Modified: Options to READ can be given before or after (or both) the
          filename.  Getting used to doing 
               read -surface s1 test.mat

Added:   WRITE <filename> -SURFACE <sname> [-BINARY | -ASCII] ...
         Similar options to normal function write.
           -SURFACE is synonymous with -CURVE and either may be used.

         write <filename>
               [-CURVE | -SURFACE <name>]      - change from default curve
               [-LIST <var1> <var2> ... /]     - use a variable list instead
               [-ASCII | -BINARY]              - default depends on setting
               [-NOTEXT]
               [-TEXT | -COMMENTS]             - prompted at write
               [-COMMANDS]
               [-COMment <comment>]            - in-line quick comment
               [-APPEND]
               [-SILENT]
               [-FORCE]                        - don't check on overwrite
               [-USER]                         - use user writeen routine

Added:   Read ASCII 3D surfaces in matrix format.  
         WARNING: This is substantial new code and likely still has some
                  errors.  Let me know of any problems.

         FORMAT is difficult, but try to be flexible:
           1. Blank lines, lines beginning with # of /* are ignored.
           2. Header section consists of all lines beginning with the
              following keywords (case insensitive):
             keyword         description                 default
           -----------------------------------------------------------
              COLS:           # of columns (x values)         <none>
              ROWS:           # of rows    (y values)         <none>
              COMMENT:        description of file             filename
              XSCALE:         spacing on x points             1.0
              XORIGIN:        x value of first column         0.0
              YSCALE:         spacing on y points             1.0
              YORIGIN:        y value of first row            0.0
              XDATA:          collection of X points (COLS)  <none>
              YDATA:          collection of Y points (ROWS)  <none>
              ZDATA:          z data begins immediately      <none>
         If ZDATA is not specified, it is assumed to begin with the first
         line of data following the headers.  Data will be read one value
         at a time in row major format.  Specificially, 
              X[0]Y[0], X[1]Y[0], X[2]Y[0] ... X[COLS]Y[0]
              X[0]Y[1], X[1]Y[1], X[2]Y[1] ... X[COLS]Y[1]
                 ....
              X[0]Y[ROWS], ...          ... X[COLS]Y[ROWS]
         Lines must be less than 1024 characters.  All punctuation is
         converted to whitespace, and values are interpreted sequencially
         independent of lines.  All values may be on a single line, or data
         may be organized one entry per line, as long as it is in the order
         specified.

         If the COLS header item is not specified, the first non-header
         line will be scanned for the number of tokens, and this will
         become COLS.  If ROWS is not specified, then file will be read
         until there is no more data and the number of completed rows will
         become the value of ROWS.  Specifying XDATA and YDATA is only
         valid after the corresponding ROWS or COLS is specified.

         If the file read options -COLS or -ROWS are specified, they are
         handled only AFTER read and interpreting, so no time is save.  All
         of the surface will be read and then will be truncated as necessary.
         This is required since the lines and rows do not necessarily align.

Added:   DIGITIZE as an internal aliased command to "READ %DIGITIZE"

Added:   "READ %CURSOR" as simple way to enter points off the screen.

Modify:  Added -NORMALIZE option (to -weighted) in the TRANSFORM Y
         histogram or cbin transformations.  Avoids the need to normalize
         to a sum of zero afterward.  If the -normalize option is given,
         the sum of all the intervals will be unity.  The command
               transf y hist 0.02 -normalize
         is equivalent to
               transf y hist 0.02
               let y = y/@sum(y)
         Also added -DENSITY to give a probability density value.  Using
         -DENSITY causes the value in each bin to be divided by the bin
         width so an integral over the curve yields the total rather than
         a sum.

Modify:  Clarified "user coordinate" setting on entrance to annote from
         RUMP.  Will now be in units of channels (NOT ENERGY) and in the
         y axis range.

Enhance: Enabled printing of full color graphs from the PM screen.  There
         are now separate palettes for printing and display (totally
         distinct).  The default page color may be specified, which will be
         used if a negative (or default) page color is specified on
         erase. (See PageColor <color> command.)

Enhance: Enable setting of page color with the command "PageColor <color>".
         For suitable devices, this will set the background page color as
         desired.  The default (or any negative value) is to disable the
         page color (unless the driver establishes some other default).
              Pagecolor <color | pen>
         color may be specified by name or as a numerical pen value.

Enhance: Annote has a new function:  FILLED_RECT <diagonal corners> <color>.
         However, only limited number of drivers support, currently only
         the PM driver.

Enhance: SUB_PLOT has new options.
              -erase
              -fill <color>
         On devices supporting filled rectangle draw, will fill the region
         of the new plot with the specified color.  -erase fills with color
         0 which should be the background color.

Enhance: Several of the annote functions now use the rectangle box cursor
         instead of having to enter two points.  Makes life somewhat easier.

Enhance: The create functions in create(), -fit, and -f options of plot now
         use separate parameter spaces.  Changes to -range, -npt, etc. in
         one form will not affect the other ones.  Also, the equation / will
         retrieve the last equation used.  These parameters are also now
         reset on a GENPLOT RESET command.  On a FIT command, the values of
         the X range will be set to the actual range used in the fitting, 
         with the number of points set to 200.  These should be good
         defaults, though can be overridden with -points or -range options.

Enhance: REGION command modified to allow all 3D values to be set at one
         time.

           region [X | Y | Z | BOTTOM | LEFT | TOP | RIGHT | AUTO | 2DRANGE | 3DRANGE]

           2DRange is equivalent to two calls with X and Y specifiers.
           3DRange is equivalent to three calls with X,Y and Z specifiers.
           AUTO sets all regions to autoscaling.

Enhance: Tired of having no flexibility in the RUMP plot command.  Added the
         following options:

            -symbol <symbol>
            -ltype <linetype>      -linetype <linetype>
            -pen <color>           -color <color>
            -symsize <inches>      -symbolsize <inches>
            -linewidth             -lw
            -l&s
            -identify <string>
            -ids
            -npoint <count>
            -shift <channels>
            -offset <counts>

         These will temporarily override the current settings.  -shift and
         -offset refer to shifts of the spectrum on the graph page.  Shift is
         specified in channels, while offset refers to a translation of the
         data in Y by the specified value, measured in units on the left axis.
         Can be used to make waterfall type plots.

Modify: In RUMP, Read/Get commands will no longer complain about invalid 
        options.  Options will fail when read as the next command instead.
        Necessitated by new options to "plot <buf>" command.

Modify: In RUMP, the Y variable and NPT are now linked to the COUNTS 
        buffer of the current data set.  Modifications using "let", however, 
        cannot be reflected in the dirty flag (no way to know they've been 
        changed).  Solely present for the very advanced user who wants to
        totally screw up their data.

Modify:  The "title" of a plot is now maintained as a static variable.  After
         setting once, you can have the same title drawn by "TITLE /".  The
         value of title is static to the routine, not the program; hence,
         GENPLOT called from RUMP will share the same title.

Modify:  If the pen is currently an autochanging value (line -1), then the
         "title" will be drawn with pen 1 rather than the current color.  This
         feature is getting painful.  Same with "identify -close" and
         "timestamp".

Add:     New transform options.  
             CENTER     - centers data set about 0.
                          Equivalent to "let y = y-((@min(y)+@max(y))/2)"
             NORMALIZE  - Gives set range of [0,1]
                          Equivalent to "let y = (y-@min(y))/(@max(y)-@min(y))"
             CNORMALIZE - Gives set range of [-1,1]
                          Equivalent to "let y = 2*(y-@min(y))/(@max(y)-@min(y))-1"

Add:     Added new option to PLOT, [-SLOW, -SLOWER, -SLOWN <ms>]
         which add a delay between points in the plotting (first time).
         This can be useful to follow order that data occur.  Slow is 30 ms,
         slower is 100 ms; slown allows the time to be specified.  This works
         in OS/2, but use in UNIX environments depends on the existence of the
         nanosleep() function which may not be implemented on all systems.

Modify:  The size of the $AXCOLR and $AXWIDTH arrays are now 6 rather than 
         the previous 5.  This is only a change in the array size published
         to the function evaulator, but not in any actual functionality.  The
         indexes correspond to:
              $AXWIDTH[0] -> axis line
              $AXWIDTH[1] -> major tick marks
              $AXWIDTH[2] -> minor tick marks
              $AXWIDTH[3] -> tick mark labels
              $AXWIDTH[4] -> title labels
              $AXWIDTH[5] -> any tertiary tick marks (log scales)
         Since these are published with the function evaluator, the "LET"
         command can be used instead of going through SGRAPH.

Clarify: Nature of TRANSFORM SQUISH, COMPRESS and AVERAGE
         These three commands may have changed slightly in definition since
         the DOS Version.  Under OS/2 and UNIX, the operations are:
             AVERAGE:  Averages <n> points into 1
             SQUISH:   Deletes <n-1> of every <n> points, tossing unused
             COMPRESS: Sums <n> points into a single point
        The use of AVERAGE maintains absolute values.  COMPRESS is
        appropriate for summing to reduce statistical noise such as in
        nuclear counting methods.  SQUISH is appropriate for random data
        sets where the density of points is just to be reduced, such as a
        random scatter plot from Monte Carlo simulations.

Enhance: 3D hidden line suppression is enabled.  Use limited to the OS/2
         PM driver and to output via the "PRINT" menu of the PM screen.
         Use "Hidden On" to enable, but be prepared later to use "hidden
         panel" to obtain the same type output.  Use the plot or overlay
         option -rainbow for limited color (to be greatly enhanced in the
         future).

Modify:  Added "axcontrol" as an alternative to axcntrl.  Also axiscontrol.
Modify:  The "DOS" command is no longer listed by the ? command.  However,
         it still works.  OS2 and OS/2 have been added as aliases for the
         shell command as well.  "OK;" is still valid also (since I'm so
         used to using that form from the days of PRIMOS).  Prefered form
         for launching a secondary shell though is !, as in UNIX, and is 
         now the listed command.  Note also that csh ans sh are also
         valid commands.
Enhance: Landscape and Portrait orientations have been extended to all 4
         possibilities.  Paper sizes defined in the device can be logically
         rotated from Landscape to Portrait, as well as 180 degrees of both.
             ORIENTation [LANDSCAPE | Portrait | INV_Portrait | INV_Landscape]
         Because of physical device characteristics, the Portrait mode
         is a 270 degree rotation with the X axis along the TOP of the page,
         not the bottom.  However, the Postscript driver does a 180 degree
         rotation on both PORTRAIT and INV_PORTRAIT plot to avoid this quirk.
Modify:  The triangle symbols (open and filled) have been moved slightly so
         that the data center is in the center of the symbol (based on the
         maximum extents - not the centroid).  This is highly debatable but
         I find it easier to discern the actual data values.  Left & right
         arrows on the same data points exactly overlap, for example.
Document: The use of ^n to switch fonts will soon be unsupported.  At the
          moment, it is backward compatible as far as is possible.  If the
          font specification "FONT HERSHEY" is invoked, then it will be very
          backwards compatible.  However, mixing new and old forms may fail
          unless enclosed in {} constructs.  
              font times
              label 1 1 '^3Time (^2l^3s)'    <--- fails
              label 1 1 '\bf Time (\mu s)'   <--- new form
              label 1 1 '^3Times ({^2l}s)'   <--- will work
          Use of old font tables 2,4,5 are especially problemsome since there 
          is no corresponding codepage/family in postscript fonts.

Document: Maximum parse size doubled to 4000*256 bytes (1 megabyte).  This
          allows an equation to have approximately 100,000 operations before
          stack overflow.  I hope it does not need to be increased again.
          The downside to this increase is that circular functions will
          take much longer to be detected.  (define f(x) = f(x)+1 eval f(1)).
          
Document: The probability of random distribution generating a given
          distribution as calculated by "fit correlate" is that determined
          from the Robust Rank-Spearman correlation.  This is statistically
          more relevent (and robust) than the traditional correlation
          coefficient.

Change: In NLSFIT, if the user does not specify error bars, they will get
        the root mean variance reported instead of the previous (meaningless
        chi-square.  Variables chisqr$ and variance$ will be linked; chisqr$
        has either reduced chi^2 or the mean variance.  The variance$ has
        either full chi^2 or the variance, depending on state of error
        specification.  Also now define variable dof$ with # of degrees of
        freedom in the fit.  List of defined variables in NLSFIT now:

                              sigma defined          no error bars
             -------------------------------------------------------------
             cf$              fit coefficients      fit coefficients
             sigma$           error bar in cf$   sensitivity factor for cf$
             dof$            degrees of freedom    degrees of freedom
             quality$ = q$   "quality of fit"             0.5
             chisqr$           reduced chi^2         mean variance
             variance$            chi^2                variance
             -------------------------------------------------------------

Document: Printing problem imbedding encapsulated postscript in Word 6.0a.
        Postscript files, including encapsulated postscript files, are by
        the Adobe standard to contain the line %%EOF as the very last line
        of the file.  This is to permit processors to detect the end of the
        real page description and avoid problems with file terminators.
        To conform to the standard, GENPLOT will always write the comment
        %%EOF to the end of all POSTSCRIPT files.  This will permit true
        conforming applications (such as Page Plus 3.0) to always succeed.
        Page Plus specifically will not import encapsulated postscript files
        without the %%EOF line.

        However, this creates a problem for interaction between some word
        processors and UNIX spooling software (specifically Microsoft Word
        6.0a and Transarc spooling filters).  The problem is actually 
        Microsoft's but their arrogance will undoubtedly cause them to deny 
        it.  When WORD imports an encapsulated postscript file, it copies most 
        of the file, including especially the document structuring comments 
        such as %%EOF, into its output file.  By the standard, Microsoft 
        should have stripped the %%EOF during copying of the EPSF file into 
        the page description.  On the UNIX side, Transarc filters are set to
        detect the conforming documents, and on conforming documents to
        send a logical END-OF-FILE to the printer after the line containing
        the %%EOF.   This has the unfortunate effect of causing the rest of
        the file to be ignored.

        In order to use Microsoft Word 6.0a, UNIX administrators must
        disable automatic EOF generation with %%EOF.  This can usually be
        done with an environment variable such as ADOBE_CONFORMING=FALSE.
        However, this may also disable the positive components of the 
        document structuring comments.  The ultimate answer is to bombard
        Microsoft technical support in the hope of getting them to make
        Word 6.0b truely conforming.

        The other alternative is to manually edit the .EPS files and strip
        the last line containing the %%EOF.  It will always be the last line.
        They may then be incorporated into Word without breaking DSC.

Enhance: At renewed interest of RdR, have gone ahead and implemented a
         non-destructive array evaluation method.  In an expression
         evaluation handled by the array evaluator, the modified array
         will be created in allocated memory, and then copied to the final
         destination.  With this change, the following will work as expected:
              let y = y-y(0)
         However, the penalty is that additional memory must be allocated to
         hold the temporary array.  For large array applications, or low
         memory situations, the in place array modification may be selected
         with the command "MathMode InPlace".  It is unselected with
         "Mathmode NotInPlace".

         WARNING: With this revision, some macros depending on this unusual
                  feature will break.  One example is the following which 
                  will sum an array with "MathMode InPlace" set.
            let y = (i==0)?y(0):(y(i-1)+y(i))
                
         WARNING: This revision only works for arrays handled and modified
                  as vectors in the code.  There *MAY* be segments of the
                  code which handle array assignment locally with single
                  calls to the function evaluation -- these will not use
                  the modified method.  I don't know of any such at the
                  moment, but the code is now sufficiently large that I
                  cannot guarentee their absence.  Let me know if any are
                  found in use.

Enhance: Added TIFF driver to the generic set.  Will output a TIFF
         B/W file at various resolutions.  Sub-drivers:
                01 - 300 DPI   
                02 - 150 DPI   
                03 - 100 DPI   
                04 -  75 DPI   
                05 -  72 DPI   
             Secondary options is bitwise additive.
                01 - draw dots as single dots, not tiny cross
                02 - suppress formfeed on output              (ignored)
                04 - suppress reset prior to output           (ignored)
                08 - suppress linewidth emulation
                16 - suppress compression algorithms
                32 - suppress blank bands at top of output
                64 - output page immediately on drawing end   (always)
               128 - rotate output page by 180 degrees
               256 - rotate output page by 90 degrees

             In general, 256 should be used when in landscape mode to get
             output presented as long edge horizontal.  Default is like
             page output with long axis down. 128 should be used if sending
             flipxy output to get output appearing normally.

             Typical entry should be 
  tiff150   tiff      02 256         01  file*query       150 DPI TIFF Output
             
Enhance: Added new option to secondary parameter of all LaserJet family
         drivers to rotate output by 90 degrees.  This option must be used
         with some care since the output will assume the printer can take
         the full "width" which is now, by default, 11 inches.  Full set of
         options to LaserJet family are:
             Secondary options is bitwise additive.
                01 - draw dots as single dots, not tiny cross
                02 - suppress formfeed on output              (ignored)
                04 - suppress reset prior to output           (ignored)
                08 - suppress linewidth emulation
                16 - suppress compression algorithms
                32 - suppress blank bands at top of output
                64 - output page immediately on drawing end   (always)
               128 - rotate output page by 180 degrees
               256 - rotate output page by 90 degrees

Enhance: MAINTAINABILITY ONLY:  Major changes to include files and some
         organization of the GENPLOT source code directory.  Will affect
         only UNIX users needing to recompile.  makefiles have been updated
         with new dependencies.

Enhance: New statistical functions in evaluator:
          @var(array) | @variance(array)
          p_chi(chi2,v)   = P(x^2|v)   = probability of chi-square, v degrees
          q_chi(chi2,v)   = Q(x^2|v)   = 1-P(x^2|v)
          t_test(t,v)     = A(t|v)     = Student t-distribution
          f_test(F,v1,v2) = Q(F|v1,v2) = F-test distribution for v1,v2 degrees
          @t_test(a1,a2)  = A(t|v) where t,v taken from a1,a2 arrays using
                            equal variance assumption with Student t-test
                            Large values ==> not likely same mean
          @u_test(a1,a2)  = A(t|v) where t,v taken from a1,a2 arrays using
                            unequal variance formulation of Student t-test
                            Large values ==> not likely same mean
          @f_test(a1,a2)  = Q(F|v1,v2) = a1 corresponds to v1 and a2 to v2.
                            1-value ==> probability that means are different.
                            Small values ==> not likely same variance
         If you don't understand these functions, it is safe to say that you
         will probably never think you need them.  (You're welcome VBS).

         The @t_test(a,b) returns the probability that a random set of samples
         from the same data set would have a t value less than or equal to
         the observed value.  99% ==> that only 1% of the time would the
         agreement be this poor.  Read @t_test as the confidence that the
         NULL hypothesis can be accepted (that the means are not equal).

         These functions precision are specified as 8 decimal digits, not the
         normal 15 digits for all other mathematical functions.  Send e-mail 
         if  you really need better precision - along with several test 
         results from primary literature.

Enhance: Added new \deg option to annote symbols.  \degree also exists.
         \degree is a character set symbol, while \deg is a manually
         tweaked superscripted o made to look reasonable good by my eye.

Enhance: Added new functions to function evaluator:
             beta(z,w)    = Beta function B(z,w)            
             lnbeta(z,w)  = Log(B(z,w))
             betai(a,b,x) = Incomplete beta function I_x(a,b)
         These functions are primarily for use with statistical tests.
         Precision for betai is only internally 8 decimal digits, though
         beta itself is full 15 digit precision of other functions.
         Better precision can be generated if necessary - send e-mail 
         along with several test values from primary literature.
         [ CHANGES 6/2011 - argument order of betai, and betai => betai_ix ]

Enhance: Added partial integration on a curve.  New function returns the
         trapezoidal integration values between specified limits.  This is
         evaluated at eval time as opposed to parse time, so creates based
         on the function are valid.
             @pintegral(cv,low,high)
         cv is a curve -- specify $plot for the current curve.
         low and high are upper and lower limits of integration.  Note that
         there is also a signed sense to the direction of x in the curve.
         Example:
            create y = cos(x) -range 0 2*pi -points 200
            archive c1
            create y = @pintegral(c1,0,x) -points 500
            plot -lt 1

Enhance: Added -avg as option for averaging on sort command.

Enhance: The characters \approx and \Approx are defined.  The uppercase
         version is bold, the lower case is normal weight.  

Change:  The < and > signs in all fonts have been replaced with glyphs much 
         more nearly matching the mathematical versions of these characters.
         Previous versions of the characters are currently available in
         font 0 as characters XY.   (Old glyphs were 2227/2228 for both.
         Replaced with 1241/1242 for normal and 2241/2242 for bold.)

Enhance: \bra and \ket are symbols for the old < and > symbol which are
         useful for creating either cyrstallographic notations or QM
         bra-ket constructs.

Enhance: Added symbols from Roger de Reus, in symbol set 0.
              Q - filled down triangle
              R - open plus
              S - open cross
              T - filled diamond
              U - open left triangle
              V - open right triangle
              W - open down trianble
         These will ultimately be accessible as normal symbols and 
         mappable into plot symbols.  As of yet, they are unnamed.

Enhance: The _{xxxx} and ^{yyy} constructs are now environment preserving.
         Font and pen changes will disappear at the closing brace. 

Enhance: A font is the combination of a family (Times-Roman, Helvetica, or one
         of the Hershey at the moment) and one or more charcteristics.  
         GENPLOT selects an appropriate font, defaulting to the Hershey 
         vector fonts when necessary.  Most of code page 850 is supported.

         The following commands select a CODE PAGE (mapping of character 
         codes to actual glyphs) without changing any other font attribute.
            font ROMAN          |    font ASCII
            font GREEK/math     |    font MATH/greek
            font SYMbols
         The following select a FONT FAMILY, resetting the code page to ROMAN
         and clearing all attributes.  This will initialize the simplest font
         in the family.
            font TIMESroman    font TIMES-roman   font TMSrmn
            font HELVetica
            font HERshey
            font SCRipt
            font GOTHic
         The following select attributes for the font.  Normal clears all
         attributes returning to the simplest font in the family.  Multiple
         attributes may be selected.
            FONT NORMAL    - clear all attributes (simple font in family)
            FONT BOLD
            FONT ITALIC

         Mapping of selected font attributes to an actual font is based on
         the above priorities.  The code page is considered first (so you
         at least get the characters you ask for!).  Next either the selected
         font family, or one with the necessary code page, is selected. 
         Finally a specific font in the family with the requested attributes
         is enabled. 

         SYMBOLS are the markers used in plotting.  GREEK contains the rest
         of the MATH/GREEK symbols.  There is only one GREEK font for either
         TimesRoman or Helvetica, and a different one for HERSHEY, GOTHIC
         or SCRIPT.

         Under OS/2, printing via the PM window should allow you to output
         with ATM fonts to any device, but I've not extensively tried it.
         Only the Postscript driver and the PM window within GENPLOT are
         enabled to draw using the ATM outline fonts.


Enhance: Even more on fonts.  The postscript re-encoding now tries to fully
         match code page 850.  If you see it on the screen, it will probably
         print to a postscript printer.  However, you must be able to enter
         the extended (upper 128) codes into the text.  Only some of the
         symbols are available as aliased \escape sequences.  Also, not all
         of these exist in the Hershey font set.  Be careful using the new
         symbols.
             eth      Eth         thorn          Thorn
             micro    section     paragraph      plusminus
             divide   multiply    degree         
             currency yen         sterling       cent       florin
         Some interesting, but not currently named, upper code page characters
             onehalf  onequarter  threequarters
 
Enhance: status will now give status of a surface as well as curves.

Enhance: plot -RAINbow on a 3D curve will give the Rainbow coloring
         effect.  This is not very clean at the moment, but at least
         a start.

Enhance: New fonts.  The system now support ATM fonts under OS/2 and
         in Postscript printers.  This is the new default (Times-Romam)
         until/unless complaints are large.  On devices without ATM
         capability, the fonts are stretched in the X dimension to mimic
         the spacing of the corresponding fonts.  The FONT (CHRSET)
         command is now changed to reflect this change also:

         See changes to Font command documented further above

Enhance: The commands 
            ov -f f(x,y) <options>
            ov -fit <options>
         now properly overlay the surface corresponding to the functions
         specified on the grid.  Options are identical to the "create -surf"
         command.  Default for range on "create -surf" are now @min/@max(x)
         and @min/@max(y).  Specifying a mesh of 0 or negative causes the
         current plot mesh value to be used.  Better choice might be the
         resolution.

Enhance: create -surface s1 -range <xmin> <xmax> <ymin> <ymax> 
         is new option.

Enhance: fit surface [-fit options]
           Determines the best second order surface fitting the current
           3D data.  a_0*x^2 + a_1*y^2 + a_2*x*y + a_3*x + a_4*y + a_5
         fit plane [-fit options]
           Determines the best first order surface fitting the current
           3D data.  a_0*x + a_1*y + a_2

Enhance: The function "3d_grid" is now implemented.  This command takes the
         current (or -curve <cv>) curve and tries to create a reasonable
         surface over the data.  This is the equivalent of the "grid" command
         from DOS 3D drawing.  The format requires the name of the surface
         to be created; other parameters are given as optoins.  Options are
         not sticky and are reset to defaults on each call.
            3d_grid <surface_name> [-options]
              ncol => # of constant X lines
              nrow => # of constant Y lines
              Options:
                -Mesh <ncol> <nrow>  
                     If not specified, current plotting mesh will be used
                -Range <xmin> <xmax> <ymin> <ymax>
                     If not specified, range taken from data set
                -MINpoints <min_pts>
                -RADius <radius>
                -Power  <power>
                -SMooth <smooth>
                -NOFit
         The algorithm is a modified version of typical gridding functions.
         The Z value is a determined by points in the neighborhood of the
         grid intersection.  The points included are determined by all points
         within a given -RADius (default 0.1), with a minimum of -MINpoints
         always include (default is 10).  In addition, more points will be
         added to ensure that there is a least one point in each of the
         four quadrants around the intersetion.  There is no upper limit to
         the number of points included in each determination.

         The points included above are weighted by their distance from the
         test point.  This weighting is set by:
             weight = [distance^2+smooth^2]^-(power/2)
         where -Power defaults to 2 and -Smooth to 0.01.  All distances are
         measured on a normalized grid of unity length in X and Y.

         If there are at least 7 points in the collection, and -NOFit has
         not been set, then the data points are fit to a quadratic surface
         and the value at the test point is determined.  The weightings are
         applied to each point in the fit.  For fewer than 7 points, or if
         -NOFit is specified, then the estimated Z is just the weighted
         average Z of the points in the collection.

         Smoothing causes points within approximately -Smooth to be weighted
         equally, avoiding huge peaks from single points near a grid 
         intersection.  A power of 0 weights all points within the radius 
         equally.  1 weights by geometric distance while, 2 weights close
         points more strongly. 

          2d create y = 0 -points 500 let x,y = 30*rnd()-15 30*rnd()-15
          3d let z = sin(sqrt(x*x+y*y))/sqrt(x*x+y*y+1E-8)
          mesh 51 51
          3d_grid s1 -rad 0.12 -smooth 0.015 -min 10 -range -14 14 -14 14
          plot s1 ov -sym 3 -pen 2

         Don't expect to get fabulous surfaces from limited data.  To
         represent the surface fairly, you need considerable data.  Or, you
         should fit the existing data to a mathematical model of the surface
         using the 3-dimensional NLSFIT.  However, this model is significantly
         better than any I've seen detailed or used.  Find a fast machine and
         be prepared to twiddle the parameters to get the best surface.
         Generally, an increasing radius will smooth the surface but will
         require more time.

         This code segment is not terribly optimized for speed since gridding
         of data should not be a common exercise.  But it is much more robust
         than the version in DOS GENPLOT which, while clever, barely worked.

Enhance: The strings describing the columns of a binary data file are now
         available for user inspection as the internal variables
         $X_descriptor, $Y_descriptor and $Z_descriptor.  These same strings
         will be written as the column descriptors on a binary write.  Since
         they have no internal purpose, it is the user's responsibility to
         ensure or ignore their meaning and usage.  Maximum string length is
         80 characters.  Strings are not local to each curve and are lost on
         the next read.

Enhance: Dealing with binary data file formatting is greatly relaxed.  The
         code now will handle most reasonable attempts to write the record
         oriented format, independent of byte ordering.  Specifically, byte
         order nad record length format are determined on the fly.  Order
         can be BIG_ENDIAN or LITTLE_ENDIAN.  Record length can be either
         the byte count of the data only (normal UNIX form) or byte count
         of data + the record markers (count+8).  Record marker must be 4
         bytes in length and must both preceed and follow the actual data.
         Also, large records no longer need to be be broken up into maximum
         of 4096 byte chunks.  Data blocks can be written as single records
         of arbitrary size.  Code itself continues to write the more strict
         format compatible with old DOS GENPLOT.  See also fix on reading
         binary files via pipe mode.  Now fully operational.

Enhance: History commands.  Behaves more like the UNIX shell history command.
         "history" lists the previous 24 command lines.  Can access (but not
         edit) previous lines similar to UNIX.
           HISTory  - lists the previous 24 saved command lines (TTY input)
           !21      - re-executes command line 21
           !cr      - re-executes last command starting with "cr"
         The comparison of strings is case insensitive but must match exactly
         the beginning of the line (excluding whitespace).  Numerical counts
         must consist solely of digits base 10.

Enhance: (MAYBE MORE) Added support for code page 850 (the international
         characters).  Nominally, the entire code page could be supported,
         but currently only a portion is.  WARNING: hdata.chr is no longer
         compatible with the DOS version!  It has an extra 160 symbols in
         each set to handle foreign symbols.  These are defined by the 850
         code page and will be created as there is help available.  If you
         want to volunteer to generate a portion of the 850 code page for one
         of the character sets, contact mot1@cornell.edu for info.

         TeX sequences to obtain without typing characters directly:

           \"a \"e \"i \"o \"u \"y \"A \"E \"I \"O \"U
           \'a \'e \'i \'o \'u \'y \'A \'E \'I \'O \'U \'Y
           \^a \^e \^i \^o \^u     \^A \^E \^I \^O \^U
           \`a \`e \`i \`o \`u     \`A \`E \`I \`O \`U
           \~a     \~o             \~A         \~O      \~n \~N

           \ss \ae \AE \aa \AA \o \O  ?` !`

Enhance: Minor modification to xdisplay.c for both XGENPLOT and XRUMP to
         allow use of <ALT><keypad> to enter special characters.  This
         allows even English keyboards to access the upper 128 characters
         of the 850 code page.  <ALT><129> (typing 129 on keypad) will get
         you the u umlaut now.

Enhance: Added one more character set to hdata.chr.  This is the triplex
         symbols in normal English only, and is symbol set 10.  Makes
         very bold symbols.  The string \bbf will switch to the bold bold
         font.

Enhance: (MAYBE) Modified the clipboard, metafile and print rendering from
         the PM screen to use only LOMETRIC resolution (.1 mm pixels) instead
         of high metric (0.01mm pixels).  This is 254 DPI versus 2540 DPI.
         This gives the plot a more reasonable size when pasted into documents
         and makes it possible to actually find it on the clipboard viewer.
         Let me know if there are problems with this lower resolution.

Enhance: Linked variables giving size of plot, margins and offset.  These
         are read only since interact too much with other components of
         system.  But at least query may be useful.
            $SIZE    - array of 2 elements (x,y)
            $MARGIN  - array of 2 elements (x,y)
            $OFFSET  - array of 2 elements (x,y)
            $SCALING - inverse of the shrink factor

Enhance: Format of the %c and %i counters in a FOR or REPEAT loop may be
         set manually.  The internal variable is $FORFORMAT and is initially
         set to "%.3i".  RESET will return it to this format.  Format MUST
         be in C format however.  Strongly recommend that it remain a %d or
         %i format just so one of the two can be used as indices in arrays.

Enhance: Add -BOX as option to CURSOR command in GENPLOT.  Final values are
         linked to the variables xbox$[0],xbox$[1] and ybox$[].  Attempts to 
         handle coherently even with non-linear axes.  Also, use of box
         cursor in most other analytical functions will set the XBOX and 
         YBOX values (fit, cull but not zoom for example).

Enhance: SAMPLE now has several new commands which permit draw squiggle
         samples in arbitrary angles.  Try ANNOTATE SAMPLE DRAW and
         ANNOTATE SAMPLE 3DRAW to get the format.  ROTSAMPLE has been
         removed since superceeded by new features.  New LABEL command
         also to put text in layers of the sample.
           DRAW  [<x_face_center> <y_face_center> | / ] <angle>
                Draws sample with center of face at specified position with
                size as previously specified.
           3DRAW [<x,y_face_edge> <x,y_face_edge> <x,y_back> | / ]
                Draws sample with face across first two points and length to
                reach to third point.
           LABEL <frac> "text"
                Draws text centered at the fraction of distance on squigg.
           ROTSAMPLE
                Prints error message suggesting use with 90 degree angle
         Other commands continue in compatibility mode

Enhance: New commands &xcursor, &ycursor, &xycursor, &boxcursor can be
         used on command lines.  Note that these are fragile commands and
         values returned depend on the last coordinates set to the graphics
         window.  GENPLOT and ANNOTE now have new commands "SETRANGE" which
         ensure the process is in a know state.
             setv xc,yc = &xycursor
             echo &boxcursor
         will pull up a cursor and return x,y coordinates which are just
         strings which may be manipulated.  See other &getarg commands for
         similar behavior.

Enhance: Pentium bug in floating point math is now detected and an error
         message is printed.  To eliminate the message, add the line
            set IGNORE_PENTIUM_FPU_BUG=YES
         in CONFIG.SYS or .cshrc (or .login).

Enhance: Both RUMP and GENPLOT now accept drag and drop of files onto the
         program.  If argument is 
             .dat or .data file in GENPLOT  --> read and plot
             .rbs or .fres file in RUMP     --> read and plot
             .hcp in either                 --> start and hc plot
             otherwise                      --> execute as a macro
         Initial directory will also be attached to the point of the
         specified file.

Enhance: Allow mapping of symbol number to symbol to change
         SYMBOLMAP <sym #> <symbol> <scaling> [...] -1
           <sym #> is number used for selection
           <symbol> is either original # of a symbol name (valid list)
              FilledSquare, OpenSquare, etc.
           <scaling> is a real scaling factor applied to each symbol -
              can be used to increase size of solid symbols so comparable
              to the open symbols.
        Settings for symbol map are universal and are only reset if you
        do a "reset -full".  Add the SYMBOLMAP to your GENPLOT.INI to get
        better ordering of symbols which having chance to remain compatible
        with old formats.

        Recommendation in genplot.ini:
          SYMBOLMAP 1 FilledTriangle      1.25  2 FilledSquare        1.25 \
                    3 FilledStar          1.25  4 FilledCircle        1.25 \
                    5 OpenTriangle        1.00  6 OpenSquare          1.00 \
                    7 OpenDiamond         1.00  8 OpenStar            1.00 \
                    9 OpenCircle          1.00 10 Asterix             1.00 \
                   11 Cross               1.00 12 X                   1.00 \
                   13 FilledLeftTriangle  1.00 14 FilledRightTriangle 1.00
                   -1

Enhance: SYMBOL <x>
        <x> can now be a name for a specific symbol.  Autochanging symbols
        must still use the integer format.  Valid names are:
          OpenSquare   OpenCircle   OpenTriangle   OpenDiamond  OpenStar
          Cross        X            Asterix        StarOfDavid
          FilledSquare FilledCircle FilledTriangle              FilledStar
          FilledLeftTriangle        FilledRightTriangle         

          Circle Square Triangle Star    (alias for filled version)
          Diamond                        (currently alias for open version)

Modify: Added COLORMAP command.
          COLORMap [ <pen #> <colorname> [ <pen #> <colorname> ] ... ] /
        The pen # are redefined to the desired colors.  Allowed names are
        white, black, red, green, blue, magenta, cyan, yellow, darkred,
        darkgreen, darkblue, darkmagenta, darkcyan, darkyellow, darkgray,
        and palegray (along with some synonyms).  The colormap will remain
        through "RESET" commands unless you use "RESET -FULL".

        WARNING: Multiple pens can be assigned to the same color and hence
                 some colors may not be available with the PEN <name> command.

        Interaction of color map and devices.  For devices where the colors
        can be controlled by the driver, such as the PaintJet or Postscript,
        the mapping will properly change colors.  In contrast, for plotters
        where the colors are set by the physical pens, the mapping is
        disabled - pen 1 will always refer to the first physical pen
        independent of the color assigned by the color map command.  This
        allows the colormap command to cause screen colors to match the
        colors that will be obtained on a physical plotter as configured.

Enhance: New ANNOTE draw command as per RDR request
            POint [<x> <y> | /] <symbol>
         will draw a symbol at the specified point.  Size is as set for
         labels.  Cannot do individual pixels though - 0 is real symbol.
        
Clarify: To do
             declare PenLt "-pen 2 -lt 4"
             ov c1 %PenLt%
         use instead
             cmdlin "ov c1 %PenLt%"
         This will work quite generally.  [Request via WLB]

Enhance: Changed -cursor to a box cursor for LSQFIT

Enhance: Added "pseudo commands" @end and @skip <n> to ASCII read in GENPLOT.
         Encountering the string @end is equivalent to the end of file.
         @skip causes the next token to be interpreted as a simple integer
         (no expressions allowed) and then skips that many lines in the file
         before continuing the read.  Both @end and @skip are case
         insensitive and may still be interpreted if you add characters to
         end.

Enhance: Added "-NOWarning" message to ASCII read to suppress all failed line
         warnings from the READ.  Error messages may still be issued by the
         interpreter trying to perform and invalid function.

Enhance: Added -YRANGE to all FIT commands.  Now have
             -XRANGE (or -RANGE synonymously)
             -YRANGE                                                  NEW
             -CURSOR (puts up box cursor and selects allowed window)  NEW
             -WEIGHT (may be ignored by some fit routines)
             -SIGMA  (may be ignored by some fit routines)
             -SILENT (may be ignored by some fit routines)
         fit <type> <type_opts> [-XRANGE <xmin> <xmax>] [-YRANGE <> <>]
                                [-CURSOR] [-WEIGHT] [-SIGMA] [-SILENT]
         on some.

Enhance: Added new command "symbolsize" as alias to "symsize"

Enhance: Added new command "symsize" or "symbolsize" to RUMP.  Sets the size
         of the symbols on plots as did sgraph previously.  Symbol size has
         moved to RUMP from general graphics and hence may operate slightly
         different.  Also, removing from common block will break old HCOPY
         files (sorry -- but officially this is still beta).

Enhance suggestion: Rather than saving images as .hcp files for long term,
                    use the PM metafile format from the graph.  IBM is more
                    likely to retain long term compatability for a simple
                    graphics image format than the more complex .hcp.  Only
                    advantage for .hcp is size -- but .met files can be
                    inserted into other graphics programs (word processors).

Enhance: Allow alternate forms as per RDR request
           el [-height <counts>] <element>
           el <element> [-height <counts>]
         in RUMP.

Enhance: Added "LINETYPE" as an alias for "LTYPE" in GENPLOT.

Enhance: Added $GENPLOT as level indicator in GENPLOT.  Use instead of
         $VERSION for future macros.  This then is consistent across 
         both RUMP and GENPLOT -- $RUMP for rump (4), $GENPLOT for genplot
         (2).  (spirit of request for $VERSION in RUMP via RDR).

Enhance: New "device" special name -CHECk which means initialize to GTERM
         variable unless there is a device already initialized.
             DEVice <name>  - initialize device by name
             DEVice auto    - initialize device given by GTERM variable
             DEVice -CHECK  - initialize by GTERM unless already initialized

Enhance: Added block comments (per RDR request)
            /** begin block comment
                end block comment **/
         All text from the beginning /** to the LINE with the **/ will
         be ignored.  The **/ can be combined with other characters -- but
         text on the same line as the **/ will never be processed.

Enhance: New command under OS/2 to both change directory and disks 
         simultaneously.
             DCD <dir>      - if of form c:\junk, does c: and then cd \junk.
         Under UNIX, DCD is identical to CD

Enhance: Invokation sequence changed to support automatic execution of
         command, data or hcopy files.
           (1) Options SHOULD begin with a -, the / is still acceptable but
               stongly discouraged since confused with valid filenames.  Due
               to validity of /help, /quiet, /inifile, ..., files in (2)
               cannot be of these forms.
                 -Help          | -?             Print brief help on usage
                 -Quiet                          Bypass login messages
                 -Inifile <file>                 Set initialization file
                 -Buffer <npt> | -BUFSize <npt>  Initial buffer size (points)
                 -3D                             Initialize in 3D mode
           (2) After all options, next token is examined
               (a) Valid simple filename with .dat or .data extension
                   "read <file> plot" executed
               (b) Valid simple filename with .hcp extension
                   "dev auto hcopy plot <file>"
               (c) Any other valid filename
                   "xeq <file>"
               (d) Other 
                   {push all remaining arguments as simple commands}

Enhance: Added new abbreviation "lv" for "listvar".  Tired of typing.

Enhance: Modified deallocate command as per request DPB.  Now
             dealloc -all
             dealloc var1[,var2,...]

Enhance: Finished ALIAS command options, especiall show and delete
            alias <name> <expansion>
            alias -add <name> <expansion>
            alias -delete [-all | <name>]
            alias -show <name>
            alias -list
            alias -execute <command>

Enhance: Reduced minimum abbreviation for TRANFSORM from TRANS to TR.
         Got tired of typing.

Enhance: Added the @absavg() function - just forgotten when others added.
         Now have @absmin, @absmax, @abssum, @absavg.

Enhance: Per request from RDR, added new internal variable $LNZERO.  This
         REAL constant will be added to the actual argument of ln() and
         log() and can be used to avoid domain errors.  Initially, $LNZERO
         is set to 0 and has no effect.  Setting it to 1E-36 will avoid
         zero crash.
            GENPLOT: eval ln(0)
            Math exception in log(): Argument singularity (you figure it out)
            Unable to evaluate expression: ln(0)
            GENPLOT: let $lnzero = real_min
            GENPLOT: eval ln(0)              /* Really ln(0+$lnzero) */
            Value:      -708.39642 = ln(0)   /* No domain error      */
            GENPLOT: let y = ln(abs(y))      /* Guarenteed safe now */
            GENPLOT:

Enhance: Added two new constants to function evaluator, REAL_MIN and
         REAL_MAX which are set to the minimum and maximum absolute values
         that a (stored) real number may take.  In double precision, 
         REAL_MIN is 2.2E-308 and REAL_MAX is 1.8E+308 roughly.  REAL_MIN is
         the smallest value that retains all the significant digits; smaller
         values may be stored denormalized.

Enhance: Occasional there will be a build of the double precision version
         of RUMP/GENPLOT.  It will be distributed as a separate name (maybe
         gdiste_d.zip).  Only one version may be loaded at a time due to
         current name conflicts.  If you have need for the double precision
         version, let me know and I will be more consistent in generating
         the files.

Enhance: Code changes so single switch in mytypes.h allows REAL to be
         become double instead of float.  This is ongoing since there will
         be continuing problems identified with code hard wired to one
         or the other.

Enhance: Changed primary name of SMOOTH_FFT to FFT_SMOOTH in transform -
         just a little more obvious and fewer letters.  SMOOTH_FFT still
         an internal alias.

Enhance: Permit "create -curve c1 y = cos(x)" to work even if c1 does not
         exist.

Enhance: Increased maximum size of axis titles and Title to 255 characters.
         Tired of running out of space for long titles.

Enhance: Function section of the on-line manual extensively updated.  The
         fit correlate has also been documented in the on-line help (OS/2).

Enhance: New "fit correlate" function to return test of linear correlation
         between x and y.  Returns traditional correlation coefficients
         (and as the variable "correlation$").  Also does a robust
         non-parametric rank correlation test using the Spearman Rank-Order
         method, returning the corresponding correlation coefficient and
         probability that a random distribution would produce a similar
         sum squared difference of ranks.  Creates new variables
              correlation$      - traditional correlation coefficient
              rank_correlation$ - robust Spearman Rank-Order coefficient
              P_random$         - probability of rank_correl for random dist

Enhance: New sort options [-x | -y | -z] changes the critical variable used
         in the sort.  "sort -y" equivalent to "exch sort exch"

Enhance: added new capability to function evaluator - single expression
         curve calculations.  Currently most useful will be the integral
         expression.
            @integral(curve)  - returns simple integral of curve specified
            @correlate(curve) - returns correlation coefficient of curve

Enhance: added new functions to evaluator
            poisson(z,x0)
            binomial(x,n,p)
            lorentz(z,x0,width)
            @median(ar)
            @mean(ar)
         Modified gauss to fully support complex mode.  poisson, lorentz
         and gauss are sufficiently well defined to permit full complex
         arguments - obvious extension to standard definition.

Enhance: On listv and savevar, allow options in UNIX style.  
            "listv -hr" replaces "listv -hidden -real"
         Because of conflicts in first letters, complex is given the
         single letter 'z' and strings use 't' for text.  -text is now
         also allowed as option in listv. Change useful primary for experts.

Enhance: Allow format "ALLOC 2D CURVE 80" and "ALLOC 3D CURVE 90" (curve is
         allowed but has no effect).

Enhance: XGenplot and XRUMP on OS/2 now have configuration windows for the
         initial number of lines on the screen.  Also can override shell
         positioning to have window start at same location always.

Enhance: On both RUMP and GENPLOT, additional line gives the Revision Date
         on the startup message.  This is the last time the original code
         was modified and hence is a good measure of the "up-to-dateness"
         of the code.  Link-time failes on UNIX machines.

Enhance: Modified code to handle shell ~<name> more consistently and
         extensively.  In any place a filename is expected, the ~user/dir
         and ~/dir constructs should be valid in UNIX.  Please send me
         reports of anywhere this fails unexpectedly.  MODIFIED AGAIN
         9/17/94 to improve further.  Should be fully supported now.

Enhance: Added ability to run pipes for read/write as per code suggest by LRD.
            read "| zcat data.file.gz"
         will read the standard output of the zcat command.   Likewise,
            write "| tee data.out | wc"
         will count the number of lines and characters in the output stream.
         Any valid command line may be given for the input.  Binary reads must
         specify -BINARY with pipe mode because PIPE is not a seekable stream.
         In binary mode, old UNIX format files cannot be read.

Enhance: Additional code to support the WATCOM 10 compiler.  Possible that
         this will compile under DOS and NT also, but don't know yet.

Enhance: Modified exception handling in keyboard input routines to better
         manage input from UNIX systems.  Should be clean now.

Enhance: After repeated abuse by Kramer types, allow modification of handling
         clipping of symbols in plot area.  As per AIP standards, no part of
         the graph extends outside the plot box by default.  This can be
         modified so symbols starting within the plot area are allowed to
         draw outside the plot area.
             CLIPSYMbols [YES | NO]
         Default is YES, set to NO to draw full symbols always.

Enhance: Allow _ as the leading character in a variable or curve.  As in
         _tmp or _data.  Just gets to be useful.

Enhance: Start documenting the 3D stuff.

  CReate -SURFace <name> [Z =] <fnc(x,y)> [-Rows <n>] [-Cols <m>] [-Grid <n> <m>]
                         [-Xrange <xlow> <xhigh>] [-Yrange <ylow> <yhigh>]

  Surface creates are handled somewhat differently.  If the surface already
  exists, the default X and Y ranges are the current ranges.  The grid spacing
  is not adjusted if no options are specified; any options or if the surface
  does not pre-exist, the grid will be defined regularly spaced.  The defaults
  values for range and grid density, if not already existing, are the last ones
  specified.   Initially, the ranges are [0,1] with a 50x50 grid.

Enhance: Trying to trap ^C on cursor input and other device messages so
         GENPLOT does not crash.  This works with discomfort in the normal
         version of GENPLOT, but wrong trap for ^C in XGENPLOT.  More work
         is needed.

Enhance: The pull-down menu for READ, Select plot device, and HCOPY are
         now operational.  The default search list is now * instead of
         *.dat.

Modify: "EDITOR" command now aliased to EMACS.  Requires all letters!  All
        of the optional commands for emacs now recognize an & after the
        command to signify a detached session.
            EDITOR [ <filename> [&] ]

Modify: Use of the external GENPLOT or RUMP macros in NLSFIT is now properly
        handled.  The GENPLOT_MACRO <name> causes the named macro to be run
        before each function evaluation.  A simple way to fit a very complex
        function is to set the function in nlsfit to trial:y, and then write
        a macro that creates trial as a curve containing the test function
        (being sure not to modify original curve).  Macro should "return"
        from GENPLOT as if it were a macro.  
           alloc results array npt
           nlsfit                     test.mac:
           equ results                   archive c1
           genplot_macro test.mac        let y = f(y) transf y integral
           vary a / vary b /             let results = y
           fit                           retrieve c1
           plot let y = results ov       return
        RUMP_MACRO can be used to begin at RUMP level as long as GENPLOT
        was called from RUMP (GENPLOT will not demand load RUMP - sorry).

Modify: The annoying message "HCOPY file active" has been dumped.  Saving
        HCOPY files on very slow computers was useful, but speed and
        enhancement of macros makes HCOPY files less useful across sessions.

Modify: The PM driver print capabilities should now be complete.  There
        are dialogs for selecting the printer, as well as to control the
        scaling and rotation.  Since most OS/2 printer drivers simulate
        colors with gray scales, the print dialog has the option to limit
        color to the number physically present in the driver.

Modify: Search for default spooler should now be more robust in I/O 
        channel SPOOL*.

Added:  Pearson.mdl file which can be loaded via "load pearson" to add
        the Pearson VII distribution functions.  See "load pearson" printed
        information.

Modify: The default curve name is once again $PLOT instead of PLOT. This 
        is done so "OV PLOT" doesn't get interpreted as "OV -CURVE PLOT".

Document: The default command allows a new curve to be used as the default
        main curve.  Initially, the curve "$PLOT" is DEFAULT <curve>. Using
          ALLOC C1 CURVE 2048 DEFAULT C1
        changes the default curve to be C1.  This command has limited usage
        since curves are now dynamically expandable.  Previously used to
        allocate larger curves for use in DOS version.

Modify: For allocation of arrays, a hidden variable :NPT will also be
        allocated and will point to the number of valid entries in the
        array.  The :npt will NOT be allocated on GVLinkArray, only for
        GVAllocArray.  

Added:  New intrisic functions sizeof(var) and typeof(var) which returns the
        number of "items" in a variable, or the type of variable respectively.
        sizeof() operates different than the C analog returning the intrinsic
        number of items in a variable type.  For simple variables, always
        returns one.  For an array, returns the maximum number of elements the 
        array can hold (use array:npt to get number currently valid).  For 
        curve, returns number of points the curve is defined.  For a matrix,
        returns the product of the dimensions.  The typeof() operator returns
        a unique code for each type, but actual values are ill-defined.

Modify: writevars now allow options to limit to specific subset
          save_var -real -int <file> 
             or
          save_var <file> -real -int ...

Modify: CONFIG is now a sub-process.  Use "RETURN" to exit from configuration
        back to normal mode.

Added:  CONFIG DATApath <path_list>     CONFIG DPATH <path_list>
        CONFIG DATAExts <ext_list>
        These commands create a search hierarchy for data files in GENPLOT
        read commands.  The path is a ; or : delimited list (depending on
        UNIX or OS/2), with directories searched in specified order.  The
        extension list similarly specifies the potential extensions to the
        base name.  The current directory and no extension are implicitely
        considered the first element in their respective lists.  The default
        search path is empty and the extension list contains .dat and .data.

Modify: BaseMacro command --> MacroPath command.  The BASEMACRO command will
        now accept a ; or : delimited list of paths (depending on UNIX or
        OS/2) for searching macro files.  The extensions for macros are 
        simultaneously expanded to .mac, .xeq and .cmd.  The extension list
        is searched in the current directory, followed by all directories in
        the specified BaseMacro.

Modify: For &getarg and similar commands, "/" is accepted to return the
        -default response (if given).

Modify: For both SAVE_MACRO and HCOPY SAVE, the file will be moved by copying
        if the temporary file exists on a different physical disk than the
        specified new name.

Modify: NLSFIT code now has two algorithms for determining parameters for
        a best fit.  The default is the existing CURVEFIT code.  The new
        one is a surface construct based on the algorithm used in PERT of
        RUMP, referred to as LOCMIN.  At this time, it is impossible to
        predict which will be better in any particular case;  however, for
        "expensive" functions, LOCMIN may require fewer total function
        evaluations, but also one get's less rapid feedback on success.
        The LOCMIN algorithm is from Larry Doolittle of PERT fame.

        METHOD CURVEFIT - This is a local gradient approach which estimates
                          the derivatives in the neighborhood of the last
                          best guess.  Optimization to avoid getting trapped
                          in the "valleys".  Stable and robust estimator
                          requiring minimal attention by the user.
        METHOD LOCMIN   - This algorithm constructs the full Hessian and 
                          estimates the chi^2 surface from N+1 points.  The
                          points are taken over the entire volume of the
                          hyper-cube defined by the variable limits.  Not
                          nearly as well proven as CURVEFIT and requires that
                          -LIMITS switch be used when defining all variables.
                          Should be tried in "expensive" calculations where
                          reasonable limits on parameters can be estimated.

        The "NLSFIT RESET" command will reset the method to the default
        CURVEFIT algorithm.  (LOCMIN code from Larry Doolittle).

Report: 12/16/93        From: DPB              Prty:    0
Want:   Edgeworth distribution for simulating implant profiles
Modify: Added edgeworth(x,x0,sigma,skew,kurt) to function evaluator.

Report: 12/10/93        From: WLB               Prty:   0
Want:   Way to change default insert mode on command entry
Modify: Added ^] as toggle of default insert mode - will be sticky to
        subsequent command lines.  Also added (hidden) commands
        INSERT_MODE and REPLACE_MODE to set these via command mode.
        Full name required.

Report: 12/10/93        From: WLB               Prty:   0
Want:   Way to print multiple lines of text
Modify: New ANNOTE command
           MULTIline [coords] { <filename }
           text - terminated by @END or <ESC>
        This command continuously requests text and places it on the
        command line.  May be from file using the <filename redirection
        format.  @end must be in the first column of the line to be
        recognized.  Otherwise it is accepted as simple text.

Report: 11/1/93         From: various           Prty:   0
Want:   Choice of using count going 001,002,003 or 000,001,002 in for %f
        and repeat processing loops.
Modify: lexp.  Major compatibility change.  %c now counts naturally starting
        at 001 and up to <count>.  %i is the index and counts 000,001, ...
        appropriate for indexing arrays.  

Report: 10/31/93        From: DPB(?)            Prty:   0
Want:   Trial function test in NLSFIT without full fitting
        Ability to break out of fits in NLSFIT
Modify: nlsfit/curfit.  New command in NLSFIT called "TRY" which just
        returns CHISQR for current parameters.  Also, numerous break
        checks for ^C within fitting algorithm.

Report: 10/31/93        From: WB(?)             Prty:   0
Want:   Use <> and >< for != in math expressions
Modify: gvparse.c done.        

Report: 10/31/93        From: MOT               Prty:   0
Want:   Avoid typing FIT NLSFIT and TRANSF CONV every time for common items
Modify: Added internal aliasing for NLSFIT and some transform commands to
        be at command level.  These simply get translated and passed on.

Report: 10/13/93        From: RDR               Prty:   0
Want:   Shorter alias for MATHMODE, use of +/- for branch cut
Modify: system.c - Added MM as alternate command for MATHMODE.
        MATHMODE [Complex | Real | BRanch <type> | STatus | WArning <level>]
          <type>  == [ Positive | Negative | + | - ]
          <level> == [ 0 | 1 | 2 | 3 ]

Report: 10/13/93        From: PMS               Prty:   0
Want:   Using Microsoft Word, import of HPGL or WPC plots fail to recognize
        the dots in line types greater than 2.
Modify: Dots in WPC driver now drawn as very short line segments which MS
        will hopefully recognize.  Another pain in the butt from he folks
        that dropped OS/2.

Report: 8/31/93         From: DPB               Prty:   0
Want:   Autorange command to reset axes to autoranging
Modify: AUTORange [all|x|y|z|bottom|left|top|right]

Report: 8/31/93         From: DPB               Prty:   0
Want:   Ability to individually force a region (bottom, left etc.)
Modify: Changed FORCE command to accomodate the new syntax
         FORCE {bottom|left|top|right} {...} [yes | no ]
        Multiple qualifiers (bottom, left, top, right, x,y,z) can be
        specified on each FORCE line.  YES and ON are synonymous as are
        OFF and NO.  If no qualifier given, all axes are processed.
           FORCE YES
           FORCE BOT OFF
        results in top,left,right all in force mode, but bottom autoscaled.

Report: 8/1/93          From: PS                        Prty:   0
Want:   Ability to change mapping of colors
Modify: Added COLORMAP command.
          COLORMap [ <pen #> <colorname> [ <pen #> <colorname> ] ... ] /
        The pen # are redefined to the desired colors.  Allowed names are
        white, black, red, green, blue, magenta, cyan, yellow, darkred,
        darkgreen, darkblue, darkmagenta, darkcyan, darkyellow, darkgray,
        and palegray (along with some synonyms).  The colormap will remain
        through "RESET" commands unless you use "RESET -FULL".

        WARNING: Multiple pens can be assigned to the same color and hence
                 some colors may not be available with the PEN <name> command.

        Interaction of color map and devices.  For devices where the colors
        can be controlled by the driver, such as the PaintJet or Postscript,
        the mapping will properly change colors.  In contrast, for plotters
        where the colors are set by the physical pens, the mapping is
        disabled - pen 1 will always refer to the first physical pen
        independent of the color assigned by the color map command.  This
        allows the colormap command to cause screen colors to match the
        colors that will be obtained on a physical plotter as configured.

Modify: The PEN <name> and the PLOT -PEN <name> commands now accept colors
        as well as integers for the pen selection.  Same collection of colors
        as given above work.  Note these work properly even if the color map
        has exchanged pens and colors.

Want:   Ability to only partially reset - leave some parameter sticky
Modify: Added options -FULL and -PARTIAL to the RESET command.  Although they
        differ little now, gives flexibility to only partially reset in
        macro starts.  Default is -PARTIAL.  Only affected property currently
        is the pen to color mapping array which remains unchanged on -PARTIAL.

Report: 7/1/93           From: MOT                       Prty:   1
Want:   Ability to use the 600 DPI resolution of LaserJet IV
Modify: LaserJet driver enhanced to exploit capabilities of the
        PCL-5 printer language.  This gives both 600/300/200/150/100/75
        resolution for the HPLJ4, but also enables RLL compression of the
        output data.
        New device drivers:
           LJET_II  - 1=300, 2=150, 3=100, 4=75
           LJET_III - 1=300, 2=150, 3=100, 4=75
           LJET_IV  - 1=600, 2=300, 3=200, 3=100, 4=75
        The LASERJET driver remains, as well as equates of above to 
        LASERJET3 and LASERJET4.  However, the format of the DEVICES.DAT
        file makes it difficult to use these alternate names.  Use the
        LJET_xxx format instead.  Any model within the class will work.
        Sub-options are as before.  For the LJET_IV, you may also choose to
        use the POSTSCRIPT driver if so equipped.

Report: 7/1/93           From: ??                        Prty:   0
Want:   Stick drawing capability - from (x,0) to (x,y) lines
Modify: Add -STICK option to PLOT to implement as desired.

Report: often            From:   DPB                     Prty:   0
Want:   Substitution arguments in alias commands - like %i in cmd files
Modify: Alias commands may now interrogate the current command line buffer 
        for replacement arguments.  Placeholders of the form %n in the
        alias definition are replaced with the corresponding token from the
        command line.

        Certain constraints and limits apply.

           o These rules do not apply to the Console Macros F1-F20 which
             are mapped directly to the function keys (OS/2 only).
           o Any string of form %n where n is a digit from 1-9 will be
             replaced by the n'th token taken from the command line.
           o Any string of form \% will be replaced with a %, but will not
             be consider further as a replacable argument.  \%1 will leave
             %1 in the buffer, and \\%11 will leave \%11 in the buffer.
           o Only the character immediately following the % is considered
             part of the dummy position holder.  %11 is a request for the
             first token with a 1 appended immediately after it.
           o tokens do not need to be used in order, and any given
             replacement token can be used multiple times.
           o Any unused tokens in the sequence will simply be forgotten.
             In string "let y = %2", %1 is read but never used.
           o Tokens beyond the number present on the current command line
             will be replaced with the empty string.
           o Normal token parsing rules are modified to permit function
             replacement.  The whitespace characters <blank> and <tab>,
             only, are used as deliminators.  Functions cannot contain
             spaces even if enclosed in (); arguments cannot be separated
             by a comma (10,10 parses as a single token).  Quotes are
             stripped from strings when taken from the command line.

        Examples (of some that work and some that fail):
           alias do_me create %1 -ra %2 %3 -po %4
               do_me f(x,a) 10 20 200   -> create f(x,a) -ra 10 20 -po 200
               do_me f(x,a) 10,20 200   -> create f(x,a) -ra 10,20 200 -po
               do_me f( x,a ) 10 20 20  -> create f( -ra x,a ) -po 10 20 20
           alias label anno label / "%1" return
               label "hi there"         -> anno label / "hi there" return
               label hi there           -> anno label / "hi" return there
           alias label anno label / "%1 %2 %3 %4 %5 %6 %7 %8 %9" return
               label hi there puppy     -> anno "hi there puppy" return
               label a,b,c    = sin(x)  -> anno "a,b,c = sin(x)" return

        Also added the new commands:
           alias <name> <exp>            Same as -add (default)
           alias -add     <name> <exp>   Add <name> with expansion <exp>
           alias -delete  <name>         Deletes <name> (if not Console Macro)
           alias -show    <name>         Shows expansion for <name>
           alias -execute <name>         Passes <name> for normal execution
           alias -list                   Reserved to eventually list all macros

Added:  Option -Bell/-NOBell to READ %DIGITIZE command.  Causes RingBell() to
        be called each time a point is accepted from the digitizer.  This
        setting is sticky across digitize calls so only necessary on first.

Added:  time() function to return the number of seconds since 00:00:00
        Jan. 1, 1970 UTC.

Added:  srand(bits) function to initialize the random number generator.  The
        value in (bits) will be stored as an integer with the bit pattern used
        to initialize the random function rnd().  Function returns seed
        value.

Modified: Function rnd() and rand() now expect NO arguments.  This may be
          a potential incompatibility with DOS version.  But much cleaner
          since the value is unused.  Modified again to permit dummy argument
          on any function that requires no arguments.  Just avoid the problem.

Added:  Ability to write multiline comments in the WRITE function, bot for
        binary and for ascii formats.  New options
           WRITE <file> [-TEXT | -NOText | -COMMENTS | -COMMAnds]
                   ...  [-COMment <comment>]
        Note the potential problem between -COMMENT which gives a one line
        comment and the plural -COMMENTS which allows multiple lines.  This
        is unfortunate but -COMMENT has existed and do not want to eliminate
        this in-line mode.  (PS - it has also been fixed for ASCII files).

        -TEXT and -COMMENTS are synonymous.  In ASCII files, the lines are
        entered preceeded by a /* qualifier.  -COMMANDS preceeds the text with
        "C " so it is interpreted as a command when `executing' the file.
        Comments may be terminated with the EOF character or by string
        "@END" on the line by itself.  This allows automated macros to write
        strings of comments to files.

        Comments are now read from the command line interface so string
        substitution is enabled.  Will also recognize the commands 
        &encode, etc.  If the arguments are out of place in macros, heaven
        help you.

Added: Modified fprintf so allows more complicated expressions as arguments.
       Commands printf "Value of expression is: %g"  sin(x)*atan2(3,sqrt(2))

Added:  Variables $MACRONAME and $MACROPATH are now defined and contain the
        name, and full pathname, of the currently running macrofile.  Useful for
        automatically annotating graphs with the macro used to generate the plot.

Added:  Complex conjugate function conj(z)

Added:  Ability to return error code from GENPLOT exit.
          QUIT [ -Fail | -RCode <rc> ] [yes]
        Without options, GENPLOT will exit with an EXIT_SUCCESS return code
        (commonly 0).  Specifying -Fail will return EXIT_FAILURE.  Any
        integer value (0-255) may be returned using the -RC <int> form.
        Use of the return code is highly system dependent.

Added:  Conditional expressions in function evaluator.
         (a) ? (b) # (c)
        If (a) is true (greater than zero), return expression (b), otherwise
        expression (c).  Can be used to define piecewise defined functions.
         define f(x) = (x<0 ? f1(x) # x<20 ? f2(x) # x<50 ? f3(x) : f4(x))
        The C convention of (a) ? (b) : (c) will also work as long as the
        ambiguity between c1:x and (b) : (c) is avoided.  Someday, should
        consider changing to C convention of c1.x as X element of structure
        c1.

Added:  Ability to call macro before evaluating function in NLSFIT
        There are certain operations that simply cannot be handled in
        default form of NLSFIT.  These include functions where parameters
        must be determined recursively from parameters given.  NLSFIT now
        has an "EXTERNAL" command which allows a macro to be specified.
        Before each attept to evaluate the function with modified "varying"
        parameters, the macro is executed through the "SYSTEM" processor.
        The SYSTEM processor allows all the normal let commands to be used.
        WARNING:  
           (1) The macro DOES NOT assume any extension.
           (2) The variables being modified by NLSFIT must not be changed.
           (3) Do not use SETVAR to modify variables.  Use only LET.  The
               physical address of the variable may be locked into the
               function evaluation stack.
           (4) The macro is called 2*N+1 times minimum for each iteration
               of the minimization, where N is the number of varying
               parameters.  Keep the macro simple for maximum speed.
           (5) Only a subset of commands may be used in the macro.
           (6) If an error occurs in the macro, you should most likely
               abort out of the fit and resolve the problem.
        This is an extremely dangerous option to an otherwise safe command.

        EXAMPLE: This is a non-trivial example.  f(x) = b*x+c where b is the
                 root of the equation tan(x*c) = atan(c/x).  b and c are
                 positive and c is to be varied.
                 
                 define f(x) = b*x+c
                 define g(x) = tan(x*c)-atan(c/x)
                 setv c = 1
                 setv b = 1
                 solve g(x) -result b
                 create y = f(x)+0.1*ndtri(rnd(x))
                 fit nlsfit
                 equ f
                 vary c /
                 extern go.mac
                 fit

                 go.mac contains the line:  "solve g(x) -result b"

Added:  Chebyshev polynomials functions.
        tn(n,x) returns the nth Chebyshev polynomial at X
        cheby(x,array) returns the SUM(array(i)*tn(i,x))

Modified: Read/Write [-CURVE <curve>] <filename> [-CURVE <curve>] [...]
        The curve option is permitted either to preceed or follow the
        filename on READ/WRITE commands.  This is a special exception to
        the normal rule that options follow all arguments.

Modified: Device drivers recognize maximum number of pens reported in
          DEVICES.DAT.
        The number of pens info in DEVICES.DAT used to be advisory, and was
        only used for recycling of "pen -1" autochange command.  It is now
        passed to the drivers, who are obliged to conform if possible to the
        specified number.  If a physical device is configured with more pens
        than specified, the extra pens will not be used.  If configured with 
        fewer, it will use the last pen for all references beyond the 
        number actually existing.

New:    RETRIEVE <array>
        Retrieving an array causes the elements of the array to be placed
        into the Y variable, with X set to the point number (0...npt-1).

        alloc a1 array 10000 let a1 = ndtri(rnd(0))
        retrieve a1 transf bin /
        plot -bar

New:    READ/WRITE enhancements.
        read -rows <first> <last> 
             This option limits reading to the specified lines of the file.
             For ascii files, this is the true line numbers.  For binary
             files, the first and last refer to the quasi-point index.  The
             first line of the file, or the first datum, is 1.
        read -list <a> 1 <b> 2 ... /
             Much improved and cleaned up.  Will correctly set the "active"
             length of arrays to actual length.  They can be over allocated.
             Example:
                 allocate errx nptmax erry nptmax
                 read <file> -list x 1 errx 2 y 3 erry 4 /
             Both npt and length of errx,erry set to actual # points read.
        write -append
             Now generates correct error code if trying to write binary mode
        write -silent 
             New option - suppresses the "Writing: [0025]" strings
        write -list <a> <b> <c> <d>
             New option to write multiple column file.  File will be size
             of the largest array, with any invalid entries set to zero.
             Example:
                 write <file> -list x errx y erry /
             Will rewrite the file read above.

New:    Root finding capability.  Created new command, SOLVE which searches
        for zero's of a function.  Tolerably stable.  Not yet implemented for
        complex functions.
          SOLVE <func> [for <var>] <low_bound> <upper_bound>
                       [-SILent]
                       [-ITERate <max>] [-EPSilon <eps>]
                       [-RESult <var_or_array>]
        Type SOLVE -? for a help on this command (not in help yet)
        I am open to any suggestions for improved syntax/functionality.

New:    The DEVICE command will return with device unchanged if none specified.

New:    Added full support for user written .USR dynamic link libraries
        and user written .MDL modules (function enhancing).  Enabled the
        external function linking into the function evaluator.
        See USER directory under OS/2 for examples.

New:    Added limits to the NLSFIT routine -- variables may be limited
        from extending beyond known physical bounds.  Format on command
        line is 
        "vary <var> <dyda> [-RAnge <low> <high> | -LOWer <low> | -UPper <high>]
                           [-LIMit <low> <high> | -FRom  <low> | -TO    <high>]
        Default is +/- infty if not specified otherwise.
        Modified "status" format to show information more clearly.

Report: 1-16-93         From:   MOT                     Prty:   4
Error:  Need gaussian too often - time to add to basic functions
Fix:    Added functions @gauss(x,x0,sigma) where x0 is the center, and
        sigma is the standard deviation.
          gauss(x,x0,sigma) = exp(-((x-x0)/sigma)^2/2) / (sqrt(2*pi)*sigma)

Report: 1-16-93         From:   MOT                     Prty:   4
Error:  Need skew and kurtosis of a distribution
Fix:    Added functions @skew(array) and @kurt(array)
        Noted that code has ability to handle array expressions.  Why don't
        they work.  Stupid author, I suspect.

Report: 1-13-93         From:   DPB                     Prty:   0
Error:  Unable to determine point where tracking cursor ended.
Fix:    Now returned as variable ICUR similar to XCUR, YCUR and CCUR.

Report: early           From:   various                 Prty:   0
Error:  Long lines difficult to handle.  Need logical continue.
Fix:    If the last character entered on a line is a \, it will be
        interpreted as a continuation character.  The character is eliminated
        and another line is requested.  Returned string is the sum of all
        continuation lines.  Works both from command line and macros.
WARNING:If you end a directory on a line, it will be interpreted as
        continue.  To avoid this, use normal "/" for all directory 
        paths inside GENPLOT.  This is in particular a problem for "cd \".

Report: 1-6-93 etc.     From:   Various                 Prty:   0
Error:  Compiler complaints from Linux and SGI Unix systems
Fix:    Minor code changes to eliminate problems

Report: 1-6-93          From:   MOT                     Prty:   1
Reqst:  Add direct support to OS/2 printers
Fix:    Added SPOOL*<queue> I/O channel.  <queue> is the name given on
        the fist page of the notebook setting for the OS/2 printer objects.

Report: 12-25-92        From:   LRT                     Prty:   4
Error:  No WordPerfect format driver (from wife)
Fix:    12-25-92
        Okay, it's back.  Device is wpdrv.  No subdevices.  Options give
        page size.  0 or 1 ==> 11"x11", 2 ==> 8.5"x11", 3 ==> 8.5"x14",
                         4 ==> 10.5"x16.5".
        Sizes will be converted to metric for the real people later.

Report: 12-10-92        From:   SR                      Prty:   0
Errors: Error function not implemented for complex functions
Fix:    12-12-92
        Yes, I need them too.
        Implemented erf(z), erfc(z) and ndtr(z) fractional error ~ 1E-8
        gamma(z)                                fractional error ~ 1E-12
        lngamma(z)                                         error ~ 1E-12

Report: 11-30-92        From:   RDR                    Prty:   0
Errors: Can't echo a blank line.
Fix:    11-30-92
        Command "echo \n" will now echo only a blank line.  This may someday
        be extended so echo can handle escaped C characters.

Report: 10-20-92        From:   RDR                    Prty:   1
Errors: Command stack only stores new commands, but does not add execution of
        prior commands to the end of the command stack.  Request implement
        something list the UNIX "history" command.
Fix:    10-24-92
        Two-fold.
        (1) Add a history command. Because of the level the command stack
            history is handled, this is a real kludge, but is done.
        (2) If the previous command to be executed is not the last on the
            stack, it will be added to the stack.  THIS IS A MAJOR CHANGE - BE
            CAREFUL.

Report: 11-01-92        From:   RDR                     Prty:   0
Errors: Wants to be able to have legend in heavy linewidth, but text light in
        the legend drawing.
Fix:    The -lw <n> option in the PLOT and OVERLAY commands is now set so that
        the legend will have the example in the <n> linewidth, but the text in
        the default linewidth.
           "lw 1 overlay <curve> -lw 3 -lt 4 -ids"
        will draw the id with lt=4 and lw=3, but text with lw = 1

Report: outstanding requests
Fix:    10-19-92
        Implemented the complex functions in the function evaluator.  Added
        new command "mathmode [complex | real | warning <n>]".  In complex
        mode, all evaluations are done with the complex evaluator.  In real
        mode, complex only used if value being saved to real variable, or if
        something in expression specifies complex (a variable, or a function
        like imag(), real(), or complex constant).  New form for ALLOC is
          ALLOC [real | integer | complex | array | c_array | string ... ]
          ALLOC [real | complex] array <n>
      * EVALUATE handles complex and real correctly, switching to complex when
        it thinks appropriate.  (eval sqrt(-1) fails, but eval sqrt(-1+0j) ok)
      * SETVAR syntax slightly changed.  Now recognizes mathmode setting, but
        also allows override if desired.  COMPLEX/REAL specifier is optional.
           SETVAR {COMPLEX REAL} name[,name,...] value [value value ...]
      * The constant sqrt(-1) is internally set using "j".  The value i is
        historically an index and cannot be changed.  However, i can be used
        in numeric constants as long it is preceeded by a digit.  1+j = 1+1i.
        sqrt(1i) = sqrt(j) = .707+.707j.  J alone may be redefined by alloc!
        It is user problem to be sure that you don't use J as an internal index
        in macros.
      * In "mathmode complex", setv switches to allocating complex constants
        by default.
      * If a complex expression is assigned to a real variable, only the real
        component is stored.  The imaginary is dismissed.
      * Functions fully implemented include all binary ops, sin,cos,tan,asin,
        acos,atan,sind,cosd,tand,asind,acosd,atand,sinh,cosh,tanh,asinh,acosh,
        atanh,ln,log,exp,pow,sqrt, relational(working w/ magnitude except for
        equalities where exact match required), chs,magn,abs,sign(on real/imag
        separately), min/max (on magnitude).  Where complex functions don't
        make sense (frac,int), only the real component is used and imag is
        dismissed.  FUNCTION NOT IMPLEMENTED bessel fncs, error fncs, ndtr,
        gamma, ln(gamma), factorial.
      * New functions added real(z) and imag(z).
      * poly(z,complex) works as expected
      * Change FFT code to alloc and set a complex array.
      * Removed the let command from NLSFIT.  LET and SETVAR now handled by
        normal common code.  (historical implementation)
      * Modified NLSFIT to accomodate complex arguments. Treated as two
        variables (real/imag).  The imaginary is identified by (imag).
      * printf and fprintf have non-C specifier 'z' now defined from complex
        printing.  %7.4z is equivalent to %7.4g%+7.4gj and will print the full
        complex value.  All other formats print REAL(x) only.
           "printf %#.7z cosh(1+j)  ==> 0.8337300+0.9888977j"
      * Allow creation of complex curves (ie. X and Y be complex arrays)????
      * Anyone volunteer to write the complex j0,erfc,gamma code?

Report: 9-28-92         From:   RDR                     Prty:   0
Fix:    10-4-92
        Modified EVAL command so will accept list of values rather than only
        a single entity.  "eval sin(pi/4),cos(pi/4),tan(pi/4)" will now work.

Report: 9-22-92         From:   JSC                     Prty:   0
Fix:    9-23-92
        Added internal variable $VERSION set to 2.0.  Can be used to test
        current GENPLOT version.  The next DOS release will have $VERSION set
        to 1.0.  For time being, you can use the format in macros, 
            SETV $VERSION $VERSION
            IF ($VERSION .GE. 2) do this or that

Report: 9-22-92         From:   EPG                     Prty:   0
Fix:    9-23-92
        Added support for color DeskJet printer.  Driver name is DESKJETC, with
        devices 0-2 being 300 DPI, 3-5 150 DPI, 6-8 100 DPI, 9-11 75 DPI. 
        Devices 0,3,6,9 allows only 1 color, devices 1,4,7,10 allows 3 colors,
        and devices 2,5,8,11 allow 7 colors.  This device is not completely
        functional; need to get hold of technical reference manual to see the
        difference between the color deskjet and the older paintjet.

Report: 9-15-92         From:   RDR                     Prty:   0
Fix:    9-16-92
        Created new REGION option WINDOW which autoranges only on the displayed
        data rather than on the entire data set.  REGION LEFT WINDOW with 
        REGION BOT 0 8 will autorange Y for data within 0<x<8.  This is a 
        first attempt, please report any problems.  PARMS display has been 
        modified to accommodate new possibilities.  If anyone has a better
        name than REGION WINDOW, I'm open to suggestions.
      
Report: 9-15-92         From:   RDR                     Prty:   0
Fix:    9-16-92
        Comments can now be terminated with a */ as in C.  This will allow
        use of comments on the end of an alias commands.  Still needed is a
        block comment structure for use in MACRO files.  The /* is still only
        recognized at the command level -- should this be moved down so 
        comments can be imbedded within a single command?  Any reason for
             READ /* this better work */ TEST.DAT
        to be allowed.

Report: unknown         From:   unknown                 Prty:   0
Fix:    04-25-92
        New command in TRANSFORM:
           TRANSForm [] CROSSing <value> [-+|-Plus|-Positive|-Minus|-Negative]
        Transform curve into X,Y coordinates of the crossing at a given value.
        Options -POSITIVE and -NEGATIVE specify particular slope desired.  Both
        slopes returned unless limited.  Returns array of estimated crossing
        coordinates with Y set to threshold.  Linear interpolation used between
        points.  Multiple points exactly at threshold return last one.  If the
        first point is equal to the threshold, it is ignored as a crossing.
    
Report: unknown         From:   unknown                 Prty:   0
Fix:    04-25-92
        New commands in CULL and TRANSFORM:
           TRANSForm AVERAGE <points>
           CULL AVERAGE <points>
        Compresses data set by averaging the x and y values with a given number
        of points.  Similar to SQUISH, but averages instead of sums the Y
        values returned, and X is set to average of X's rather than the
        mid-point X.  Final data set has (npt/points) data values.

Report: always          From:   Thompson group          Prty:   1
Error:  ls is far too simple.  Want to use options of command line version.
Fix:    04-09-92
        Implement full structure ls/ll/lf commands.
          ls [-abvdef] <spec>           (massive # of options)
          ll [-abvdef] <spec>           (massive # of options)
          lf [-abvdef] <spec>           (massive # of options)
        ls,ll behave like the unix counterparts.  lf is specific to OS/2.
        Options include sort criteria (-date, -size) as well as listing
        formats (-l, -f, -a, -d) as in UNIX.

Report: always          From:   Thompson                Prty:   1
Error:  The @fncs such @sum and @ave only accept arrays currently. Would like
        to accept arbitrary expressions instead.
Fix:    04-07-92
        Modified gvcalc() code to handle arbitrary expressions in the @fnc's
        during parse time.  This also speeds up operations.  @max(y*x) will
        search correctly the function space y-x.

Report: always          From:   Thompson                Prty:   1
Error:  Need a coherent way to set the IDS of a file read
Fix:    27-Feb-92
        Added code to keep first line of comment in data files as the IDS on a
        read. IDS first set to filename (as previously), but if any comments
        are found in the file read, the IDS is changed to that string.  On
        writes, inverse is also performed.  If IDS is non-blank, first line in
        file will be a comment of that string.  Implemented on both ASCII and
        BINARY reads/writes.  To distinguish between real comments and
        filename, filename is prefaced with a !.  If the first character of an
        IDS is a !, comment writing is disabled.

Report: always          From:   Thompson                Prty:   1
Error:  Why can't ALLOCATE command in SYSTEM take a list of variables
Fix:    29-Feb-92
        Okay, let it be a list.  Allow the format
           ALLOC a0,a1,a2,a3,a4 REAL
        where all will be allocated as REAL's

Report: always          From:   Thompson                Prty:   1
Error:  Need much more flexibility in the outputting/formatting of text
Fix:    29-Feb-92
        Added commands PRINTF and FPRINTF to the SYSTEM menu.  These work like
        their C counterparts, with printf output to the screen and fprintf
        output to the script LOG file only.
            printf  "a0: %12.4f  a1: %12.4f  chisqr: %g" a0, a1, chisqr$
            fprintf "a0: %12.4f  a1: %12.4f  chisqr: %g" a0, a1, chisqr$