LIST OF ERROR FIXES IN ANSI-C DEVELOPMENT
===========================================================================
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

===========================================================================
QUICKIE MOD REPORTS:
===========================================================================
8/7/2013: let could not set individual elements of an integer array or
   a linked integer array.  Not a problem until the $AXCOLR was changed
   from a real array to an integer array.

6/13/2013: Fixed problem with binning faulting in error if all the
   values were equal.  Had to do with trying to round 0.00 to a fixed
   number of significant digits (duh ...).  Sometimes we are not so
   bright.

2/14/2013: *VERY OLD* bug squashed.  When printing a graph at less than
   100% scaling, the width of the lines increased rather than decrease.
   Was dividing by the scaling factor rather than multiplying.  Ouch ...
   should have looked and fixed this *long* ago.

2/8/2013: Fixed read so that one can use options to %digitize again. Once
   axes are calibrated "read %digitize -hold" will once again give you
   the ability to reuse same coordinate system.

1/17/2013: Minor fix after stck problem.  Moved block code "holding" the
   stack pointers before function calls to surround all function type
   calls in GVPARSE.  Problem showed up with GVLINK'd functions from the
   GPIB module.  Need to keep eye on other types for functions.

1/2/2013: Fixed strange problem that's been hanging around "function"
   evaluation functions such as sum(), prod() and integral.  Showed up
   first in a solve() where the second version of this code failed.
     define f(x,y) = x+y
     eval f(solve(x-1,0,2),3)   /* succeeded
     eval f(3,solve(x-1,0,2))   /* failed
     eval f(sum(i,0,5),5)       /* correct
     eval f(5,sum(i,0,5))       /* incorrect
   Finally tracked the problem down to stack referencing in the parse
   routine.  Should be resolved now.  I tried to test, and hopefully
   have done enough to expose obvious problems.
     eval f(solve(x-1.5,0,solve(x-2,0,3)),solve(x-0.5,0,2),solve(x-1,0,2))
   But as a result, added good debug code to math operations.  For fun,
     let $SysDebugFlag = 6
   and then do some math operations.

1/1/2013: Fixed problems with wait.  Now functions properly with time out.

1/1/2013: Fixed problem with int2bin(2^32).  Now works with any value
up to 1E19 roughly (64 bit precision).  Also modified most of the
evaluation routines to work with 64-bit integers so won't have problem
again.

11/26/2012: The "define f(x) = 1.02x" failed to generate errors and did
   not do the apparent implicit multiplication.  Problem is the function
   variable substitution changes "1.02x" to "1.02&00" which was then
   interpreted as a binary AND operation.  Changed the & in the substitution 
   to the character 0x7F so that this will not happen anymore.

10/23/2012: Fixed occasionaly problem with @median() function that would
   fail to converge to the median.

10/8/2012: Fixed overflow problem when doing convolution of data on very
   large number scale (SIMS data at 1E21).  Not sure exactly the error,
   but solved quickly by normalizing both arrays to [-1,1] and undoing
   the scaling at the end.  Modified corrl() routine in fft2c.c.

9/17/2012: Changed slope/offset language to slope/intercept in linear fits.

9/17/2012: lrand48() and mrand48() functions used lower 32 bits of the
   LCG sequence.  Changed to use upper bits which have better random
   behavior.

1/31/2012: Several CRT routines (such as close()) called an OS error handler
   for invalid parameters.  The default action of this handler is to
   terminate the program and generate a crash report.  This is annoying
   in evaluation where such errors are common.  Used CRT function
   _set_invalid_parameter_handler() to cause these parameter range
   errors to be ignored.  Functions just return an error rather than
   crashing.  Debug mode can be used with the handler to figure out
   where the failures occured.

12/6/2011: Switching arguments in BETAI_Ix did not get fully propagated
   to all uses.  The @t_test(ar1,ar2) for example failed.  Nopefully now
   all internal uses are again consistent.

11/7/2011: Fixed problem allocating a empty string
    Reported segment error with command
        declare astr = ""
    failed.  Problem with initial allocation of strings now NULL instead
    of a default length.

9/3/2011: Fixed problem with logarithmic axes that were more than 7
    decades and all values were negative

8/28/2011: sprintf function did not accept "C-language" strings with
   embedded control characters.  Now fixed ... printf, errprintf, and
   sprintf all work the same.

7/16/2011: maxtrix insert_data / extract_data
   Both ROW and COLUMN executed as column, and both X and Y executed as
   FOW.  Fixed both commands so can do both column and row.

7/15/2011: matrix extract
   Off by 1 in both rows and columns.  Problem with 1-based indexing for
   rows and columns versus 0 based indexing in code.

3/11/2011: pwd() and cwd() got wrong argument list in code change for
   arbitrary argument structure.  Complained of too few arguments.  Simple fix.

3/10/2011: Still problems parsing when E is the dummy variable.
              define f(e) = 1.07E-7*E-3
was failing since there E-3 looked an awful like 1.07E-7.  Had to
add check that character preceeding was not a decimal digit or decimal
point.

1/5/2010: sprintf() could not accept a string replacable argument
             define f(#text) = sprintf("a%sb", text)
          failed.  Now should work

10/10/2010: binomial(x,n,p) modified to return 0 for x<0 or x>n

9/20/2010: Minor change to @skew() and @kurt() to return proper
estimator of parent moments in case of small skew/kurt and small sample
number.  Now done correctly (or at least as most other definitions
implement these moments).

9/x/2010: Fix ls so "ls adir/" succeeds as if "ls adir" rather than fail.

7/27/2010: isatof() and isatoi() now return FALSE (0) for an empty string.

5/17/2010: Changed the flow of the GENPLOT read to fail gracefully with
     bad options.  Attempting to do a binary bitmap read from the
     console now just gracefully gives an error message.

5/17/2010: Change to allow recursion within function of functions.  Now
     valid
            solve(integral(t^2|t,0,x)-1|x,0,2)

4/19/2010: 0j+SQRT(-1) returned -j instead of j!!!  Windows compiler
     seems to randomly change between +/-0 and distinguish the two.
     Added line "if (z.y == 0) z.y = 0" which fixed the problem :-(

3/16/2010: Fixed minor problem with "click" start of XGenplot or Genplot.
     Was failing to terminate the string from command line options.  Now
     should behave better if you associate .mac and .dat files with
     XGenplot (or .rbs files with RUMP).

2/26/2010: Ouch - very strange problem.
        define f(e) = 3*e*1E-7
     would fail because the E in 1E-7 was considered a dummy variable.
     Put in essentially a specific check for this problem (gvcode).
       define f(E) = 3*E*1.0e-7 create y = f(x) -range -10 10 lt 1 pl

2/3/2009: Fixed minor problem when running shell commands in a directory
     path that contains spaces.  In XGenplot or XRump, external commands
     must be passed to the command processor to execute as if they had
     been typed at a console session.  That means quotes around the
     executable path if it contains a space.  Oops - forgot that since
     never put spaces in names normally.

10/26/2008: Selecting an invalid palette (-palette ) with plot
     would crash GENPLOT hard.  Now just defaults to some palette to
     continue running with error message

2/6/2008: The bit-wise operators & and | failed due to addition of new
     math operations.  They needed to be one of the first 255 listed and
     that condition was violated.  Now moved to the very early part of
     the list to avoid in the future.

12/11/2007: The common "-curve " modifier for commands failed to allow
     for a 3D curve.  Had "(itype != GV_2DCURVE) && (itype != GV_2DCURVE)"
     where the second one should have been GV_3DCURVE.  The implicit check
     was valid.

11/28/2007: Minor correction to the @t_test(ar1,ar2) code.  Now uses
           fully proper expressions for the pooled standard deviation.

10/7/2007: Long overdue fix to ls.  sizeof(ptr) versus strlen(ptr)

10/4/2007: One fix to string arrays broke key use.  Temporary use of a
           variable that got fully utilized in new code.  Quit kludging!

10/1/2007: Fixed problem with Rank-Spearman which appeared with last
           change to heap_sort procedures

9/20/2007: The error with DIR on huge directories has finally been
           resolved.

9/12/2007: Allocating of string arrays tied maxsize to size.  Has been
           fixed so changing size doesn't modify the allocated size.

9/12/2007: The forlist (var) failed if length of var was greater than 4096.
           This arose from a test
               if (space availble < size of var) {
                   increase space by 4096
               }
           changed so increase is at least bigger than sizeof var.
           Single items in the final list still cannot exceed PATH_MAX
           characters, currently 256.

8/30/2007: The -range command in fit commands has been failing for some
           time.  Not a problem with specification of an XRANGE, but
           always with Y.  Ouch ...

8/23/2007: Some bitmaps report negative height or width.  Now use absolute 
           values, but report as a warning.

6-2-2007:  *Very* minor fix to evaluator precedence.  Bitwise or was
           given precedence of logical or.  Doubt anyone hit that bug :-)

5-12-2006: Revised slightly the operation of the matrix read code.  The
           read will only proceed as far as the request for data extends
           from the -ROWS   option on the read command line.

3-22-2006: Fixed memory leak in matrix rotation code.  Dumb mistake :-(

10-10-2005: Put in a "GVTrimToReal" for all ASCII reads.  Reading a number
            out of range (10^309) was causing overflow exceptions.

8-2-2005:  Tired of slow main-curve expansion when reading millions of
           points.  Now number of points doubles each time an ASCII read
           must expand the main curve size.

8-2-2005:  Modified the AVI read code to permit reported image sizes that
           are greater than the product of the width*height*bytes.  Do not
           understand the underlying cause, but just choose to print a
           warning and ignore the trailing information.

6-27-2005: Fixed error reading "lookup-table" AVI files.  Returns the index
           of the table entry, not the actual color.  B/W files will range
           from 0-255 instead of 0-765 under this circumstance.

6-24-2005: Change to curfit/nlsfit.  At some point, the alpha matrix was
           normalized to improve/stabilize the matrix inversion.  Some
           data sets however do generate diagonal values of zero which
           then cause divide-by-zero error in code.  This is now explicitly
           checked in the code, turning off the normalization if zero.
           As it indicates a variable with no chi-square change, there
           is likely to remain a real problem with fitting, but at least
           the code won't hang.

11-9-2004: Forced pl -lt 0 -sym 0 to no longer hang up forever :-)

10-19-2004: Cleaned up internal code in allocate() and added help. alloc -?
            gives a reasonable explanation of how it works.

10-18-2004: Minor problem in NLSFIT.  If working with a 3D function, it only
            defined the function f(x) instead of f(x,y).  Fixed while drinking
            heavily on flight to SFO, so hope no other problems introduced.

9-20-2004: If you just start GENPLOT, create y = sin(x) will crash.  Ill defined
           limits.  All of the @xxxx(array) functions now return 0 if there is no 
           data in the array (n = 0).

7-3-2004: Finally resolved the issue of fft-filter crashing after just
          starting xgenplot (occasionally).  Failure to reset the
          pointers after increasing the curve size was at fault.

6-7-04:   Return of coordinates from cursor -box were exchanged.  Had
          xbox$[1] and ybox$[0] exchanged.

2-5-04:   Despite writing a help for RUMP WRITE command, no way to get it
          to display.  Had to revise LexCheckHelp() so that a "test" could
          be performed and have request moved from bmanip.c to rdwr.c.

9-26-03:  Fix bug in drawing of grid lines on a non-linear axis.  RD
          discovered problem of lines going off bottom of graph.  Oh well...

4-9-03:   Fix minor bug in string variable editing dialog box.  The C
          function sprintf(szTmp, "%-20s %s", name[i], adr[i]) fails at
          something like 1500 characters in adr[i].  Replaced it with a
          safer string copy and concatenation functions.  Also changed
          the string editing dialog box to AutoHScroll to allow editing
          of these rediculously long strings :-).  Have got to get
          procedures implemented soon.

4-1-03:   The addition of the code to make ^Z auto-minimize the XGENPLOT
          window broke the code for writing macro's and logfiles.  These
          commands used ^Z as the end-of-text indicator and hence failed
          with no recourse :-(  This has been fixed - the behavior of ^Z
          is now state dependent, working properly in each case as expected.

3-31-03:  Fixed annoyance with variable name translation in the presence
          of % characters in the existing strings.  Occurred in places where
          the entire command string would be interpreted as a single object,
          such as echo or ok;.  The behavior occurs with expressions such as
               echo power=54%/filename is %newname%
          The parser saw the first percent and interpreted as the beginning
          of a possible variable name.  Scanning to the first %, it looked
          for the variable name "/filename is " which - obviously - did not
          exist.  This however got everything out of sync looking for more
          names. The code now tries ignores the % pair if the string it
          would enclose contains whitespace.

3-14-03:  Trivial.  rand() now is on interval [0,1).  Value 1 is never
          actually observed.  Previously had same probability as 0.

12-15-02: TRANFORM ZERO_PAD failed to expand the size of the main curve if
          npt request was greater than nptmax.  Now does so.

11-11-02: Random number generator is now initialized to a random (time based)
          seed in RUMP as well as GENPLOT.  OOPS ...

10-10-02: The file handing had a slight problem with some names and FCLOSE.
          It was often necessary to deallocate the variable before reusing
          or risk getting hung up.  The code was changed to refer so that,
          on fclose, the variable is reset to NULL and is known to be closed.

4-29-02: Identified the source of the failure of XGenplot and XRump on Dell
         4100 laptops.  In redirect.c, reset stdio and stdout handles were
         relinked to the SetStdHandle.  Have to have STD_OUTPUT_HANDLE to
         get output from external programs to the screen (! dir).  But don't
         have to have STD_INPUT_HANDLE as long as programs do not require
         input.  If they do, we may crash.

4-18-02:  Pipe stream reads now finishes with a full purge of the pipe.  This
          eliminates the annoying "GZIP Pipe Closed" message for instance.
          Could be a problem if the pipe never shuts up, so there is also
          a limit on the total number of purges.  Minimum read is 100 kB
          but actual will depend on previous read increases in buffer size.

3-9-02:   CURSOR -TRACK gave meaningless message when npt was zero.  Modified
          error message so explains real problem.

3-4-02:   Fixed problem with fopen in pipe-driver for UNIX.  The fopen() in
          Linux is now adament and will not permit "wb" (binary) options. 
          #ifdef segment to switch between needs of UNIX and Windows.  Also
          minor fix in UNIX to null-op on VIRTUAL_SHIFT_TAB characters (only
          useful for reverse tracking filename completion in Windows).

1-28-02:  Fixed problem with color failure on PM graph screen.  Not sure of
          the exact root cause, but clearly a race condition.  Inserted
          100 ms of sleep in the thread dispatching plot commands to the
          main graph window from drvchild.  Seems to have fixed many
          of the stability problems.

9-18-01:  PLOT -PLX TOP has failed for some time.  Apparently nobody uses
          it, including me, until now.  But has been fixed.

4-30-00:  FIT POLY 0 now fits y to a constant.  Crazy way to do it, but
          makes it at least consistent.  Same as @ave(y),@std(y).

8-7-00:   Fixed problem with filled symbols in reversed color orientation on
          the PM screen.

8-2-00:   Checked for negative index setting in arrays.  c1:y[-2] will now
          print the same warning as going beyond the end of the array.

5-16-00:  Powerpoint bug squashed!!!!!  Maybe.  See comments in dialog box.
          
03-14-00: Minor fix in deallocation of command structures with error bar
          plotting.  Should not impact anything - showed up in new compiler.

02-29-00: YEAH! Maybe a fix for the ocassional window going totally crazy.
          Switch to MSC 6.0 compiler identified wrong parameters to calls
          on EndPaing and DeleteDC.  These may have been cause of resource
          overload that appeared in font selection, etc.  Fixed now.

          Bad news: There seems to be another hang introduced by MSC 6.0.
          Must be careful.

09-30-99: Made \box work from within annote.
08-13-99: Minor fix to wordpos() to handle whitespace in needle, and to 
          avoid matching start of words.  Must match full word.
07-09-99: Some problems in function parsing - involves string functions
          only.  Should be resolved.
                define f(#x) = strlen(strcat(x,"hi   "))
          would lose the spaces in the character constant.
01-14-99: Changes xcur,ycur,icur,ccur from being dynamic variables within
          the GENPLOT common block to being static variables.  The use
          of pop_state would pull these variables out from under their
          definition and lead to potential invalid memory references.
01-04-99: Dieckmann pushes the limits again.  Got me looking into ERFC
          again for extreme arguments.  Modified handling between 4.0 and
          15 or 27 (depending on machine) to have fractional error within
          1E-7.  Now uniform over range [-be with 1E-7 of correct.
          Also implemented lnerfc(x) to handle really bad arguments from
          diffusion profiles.
12-28-98: Subtle problem with functions of no arguments.  The following 
          was failing.  
            define h0 = 2
            eval 1*h0
          Tracked down and fixed -- but in midst of major overhaul of the
          function evaluation so may have other problems.
12-21-98: Spent too many hours searching for a NLSFIT problem where the 
          limits were incorrectly given in reverse order (lower > upper).
          Added code to check and spontaneously reverse - but with warning.
          At the same time, extended massively the information printed
          in mode VERBOSE 4.  Recommend not running this mode :-)
12-17-98: Request to extend ERFC function to rediculous limits.  Now will
          do to arguments of 15.065574, where ERFC(x) is 1E-100 (UNIX) or
          to 27.226 where ERFC(x) is 4.94E-324 (minimum of Pentium math
          precision).
9-16-98:  Yet yet another minor fix to @median to handle case where the 
          median is exactly equal to the maximum or minimum of the range.
          Boy, talk about bad data sets :-)
9-2-98:   @median(y) failed for wide dynamic range (1E-5 to 4000).  Modified
          so much more likely to converge, plus now gives error message if 
          it fails to converge.
8-14-98:  Radius for ANNOTE->CCIRCLE now properly registered from cursor.
8-4-98:   PAGESPLIT -? now gives help (as well as ? alone).  Also added
          SPLITPAGE as alias for PAGESPLIT.
7-10-98:  For P. Smith.  Added check in nlsfit so can be used for function
          minimization where there is perfect error correlation between all
          of the variables.  Removes the divide by zero error.
7-7-98:   Fixed word(xxx) and subword(xxx) functions to return null string
          if ask for word more than in string.
7-7-98:   Fixed subtle problem so that "substr silicon" is interpreted as
          a non-string -- looking now for () to mark function calls in the
          expression type guesser.
5-3-98:   Fixed problem with difficulty moving windows when the cursor
          was active in XGENPLOT.  Windows would have their mouse posns
          constantly changed by the cursor even when it didn't have focus.
2-20-98:  Functions of no arguments may again be given no arguments.
             time(), rnd(), etc.
2-19-98:  Modified the ROUND(x,n) function to handle exactly 0.5 better.
          Had always rounded one way -- now has modification to round on
          average up and down equally.
2-16-98:  Duh!!! A minus sign is considered punctuation.  Well what planet
          was I born on?  Matrix read now deals with negative numbers
          properly.
2-13-98:  Added skipping over comments to 3D matrix read routine.
1-20-98:  Added translation of %var% to expression in &encode usage.  Fixed
          so following works as expected.
               declare var = ids
               echo &encode "The value of %s is: %s" var %var%
1-18-98:  Improved handling of large X offsets in convolution (and related)
          routines.  Should avoid the subtle shift noticed in some convolution.
1-15-98:  Fixed argument pulling in LEXP so block statements (if blocks)
          don't prevent the use of &getarg
11-29-97: Fixed so all of the axes get background colored with pagesplit.
10-22-97: Changed the environment variable "DYNT_LIB" to "GenplotSearchPath_NT".
          This allows the setting inside the registry to be overridden at
          run time via an environment variable.
10-21-97: Finally found the resource problem in the "pipe" internal calls.
          The continual opening of handles is solved and should no longer
          cause the desktop to "hang-up" when running out of resources. 
          Forgot that CreateProcess() returned open handles to both the
          process and the main thread -- wasn't releasing the thread handle.
          Seems to have resolve the problem.
07-24-97: Fixed gptfit so that "create 0 fit poly 2" does not crash with a
          math divide by zero error.  Range of data = 0 was a problem.
06-24-97: Finally found a way around the Windows 95 crash with internal
          commands via the OK: command (I think).  An extra parameter is
          required on the command line which is now automatically generated.
06-14-97: There appear to be some conditions where @rms(array) is hitting a
          DOMAIN error on SQRT.  This appears to occur when the RMS is very
          small compared to the average of the data set.  The code has been
          changed so that if the @RMS is within 1E-4 of the absolute value of
          the average, it is computed by the direct average of <(x-)^2>
          instead of the algebraically expanded equivalent. (gvcalc.c)
05-25-97: Apparently I added the horizontal scroll for RDR in OS/2.
05-19-97: Eliminated divide-by-zero problem with dy/dx transform where data
          was not strictly changing (x[i+1] = x[i]) making derivative very
          difficult.
05-17-97: Stack overflow problem when specifying file*cmdline in XGENPLOT
          fixed.  May resolve other problems also. (OS/2 version only)
04-01-97: Minor problem - fflush() for GNU was only a flush() in con_unix.c.
03-21-97: Under Win95, VSORT.EXE (and all of the pipe drivers) failed.  Had
          to remove \LD compiler switch to force 95 compatibility.
03-20-97: Subtle error with ClipSymbols -- if an erase was not done before
          the plot with autoerase off, then the symbol no-clipping failed.
          Required adding ClipSymbol into HCOPY file structure.
03-20-97: NT: Printer dialog now retains the last used printer.  Metafile
          dialog now writes in the correct directory.
02-18-97: Screwup at Microsoft again (does this surprise me??? no.....).
          Powerpoint does not interpret graphics on the clipboard the same
          as any other Microsoft product.  Copying a graph and then pasting
          into Powerpoint screws up any letters not drawn along the 
          horizontal.  Pasting the exact same clipboard into Word, Excel,
          or the clipboard viewer works absolutely fine.  Only Powerpoint.
          The solution from Microsoft???  You should ask.  Paste it into
          Word, then copy it from Word to Powerpoint.  Great!  My solution,
          make it obvious that Microsoft is screwed up -- add a separate
          button to copy to the clipboard for Powerpoint.  This distorts the
          characters exactly as is necessary to compensate for Powerpoint's
          screwups.  Why oh why am I back fighting with Microsoft?????
02-17-97: Fixed problem with PLOT -SYM 31 - memory read buffer overrun.  Syms
          now wrap around at the appropriate point.
02-17-97: Added TrueType font support to NT PM screen driver (and printing)
02-17-97: Fix NT problem with special symbols in Postsript driver
01-29-97: Problem in NT with a particular spline.  Not sure if compiler or
          operating system or what -- fixed by increasing internal work of
          spline to double precision and carefully converting back to REAL
          via GVTrimToReal().
01-25-97: Fixed failure of CLIPSYMBOL NO in PORTRAIT mode.  Modified
          PlotInBound() to use new PlotConvert3DScale().
01-23-97: Fixed SUB_PLOT erasing too much on NT PM screen driver.
01-23-97: Fixed introduced bug where &getarg failed - changed code of GetToken
          to safely save backup pointers.
01-17-97: Fixed minor problem which caused commands like ACTIVE alone from
          SIM in RUMP to be ignored (returned to RUMP but did not execute).
          Arose from default check for automatic variable setting.
01-03-97: Fixed GRID command in NT/95 (illegal address reference)
12-20-96: Maximum length of IDS on plots now DFLT_STR_SIZE (256) characters.
          Previos was 80.  Also won't overwrite other structures if too long.
12-16-96: SORT -STRICT on a 3D curve now looks at both X and Y to be identical
          before dropping the point.  Real screwup before.
12-16-96: "Typing" a file clears the entire buffer in OS/2 XGenplot version.
          This now is modified so it just puts out 5 blank lines before
          starting to output the file.
12-16-96: Fixed -L&S so works if common points in dataset
12-2-96:  Squigg labelling (annote squigg label  ) failed in
          PORTRAIT mode.  Now fixed.
9-16-96:  Compiler patch to CTC005 installed -- seems to fix problem with
          "sort" and "fixgrid" identified in latest version.  Who knows
          what has broken now ...
9-7-96:   Initial NT port complete.  Relatively small changes so that don't
          expect major problems cropping up in the OS/2 and UNIX versions.
8-23-96:  Fixed 3D axis problem when in other than landscape orientation.
8-23-96:  Fixed read ~mot/xxx format usage on UNIX.
8-8-96:   Documentation.  OS/2 TCP/IP screwup.  It is not possible to pipe
          binary input into the lpr function from the TCP/IP 3.0 package.
          While text can be sent via stdin, binary gets translated no matter
          what one does!  -b only works for files that are opened by lpr.

8-2-96:   Changed reading for pen colors to use math mode allowing complex
          functions such as mod(%c,7)+1 (which failed before due to the ,)
7-13-96:  The file reading routines stripped the | sign before it could
          be passed to user routines.  This is now left as part of the
          filename so can be processed elsewhere also.
               read "| mac2os2 terminal.log" -user
          can be used to read a Macintosh file in some strange format.
5-28-96:  Increased size of equation and weighting functions in NLSFIT to
          256 characters from 80 characters.
5-2-96:   Fixed possibility of memory overwrite when first comment line in
          an ASCII data file was longer than 80 characters.  Now, the
          comment is truncated before being assigned as the curve IDS.
2-11-96:  Fixed problem with Lines_And_Symbols in flipxy mode.
2-11-96:  Compiler bug causing failure of RUMP smooth bypassed with dummy
          code under CSET/2.  See anlytc.c for problem.
1-4-96:   Pentium bug check removed from executable code and placed in the
          install code instead.  Hopefully it is now robust enough to not
          falsely flag the error.  Will only be warned at install and then
          it up to the user to worry about later.
1-4-96:   Modified character handling in pmdriv so that ~C will toggle the
          cursor view window whether the focus is on the main graph window
          or the cursor window itself.
1-4-96:   Fixed the symbol definitions for \varphi in Hershey set.
12-15-95: Fixed the problem with error bars running through the center of
          open symbols.  Will again work as intended.
8-17-85:  Added tilt,rotate, etc. to parms listing in GENPLOT.
8-17-95:  Fixed failure of ANNOTE to handle user coordinate system when in
          Portrait mode.  Oh well ...
8-15-95:  Increased a number of internal string variables to 256 characters
          or more.  Resolves pain in "declare" usage with very long strings.
8-13-95:  Specifying a linewidth of less than 0.14 in the "LINEWIDTH" command
          will now set the linewidth to the minimum .001 inches.  
8-13-85:  Well, actually put in code to support the "Disable Math/Greek"
          and "No cursor display" buttons in PM Driver configuration.  Looks
          like I started to implement it and went to bed several months ago.
7-20-95:  Changed handling of temporary files slightly.  If the TMP
          variable points to an invalid directory, it now prints an error
          message and proceeds to use the current directory.  Won't trap
          on all invalid TMP names, but most.  For example, set tmp=m: will
          fail non-graciously if m: is not a real disk.  Also,
          "set tmp=e:\tmp\" is considered invalid, though "set tmp=e:\" and
          "set tmp=e:\tmp" are all valid.
7-20-95:  Changed shortcut-key for "FIT" menu to "I" so doesn't conflict.
7-20-95:  Fixed problem with ^Z in macro files (I think).  If a logical end
          of file character is seen on any line in a macro, the line is
          taken to mark the end of the macro and the line is executed only
          to that point.  The file is then considered as closed.  On OS/2,
          the logical end of file is ^Z, in UNIX, it is the ^D.
6-2-95:   3D failed to draw axes.  Problem in transfering the 2D clipping
          box (margin, offset) in a 3D environment.  FIXPLT.C modified.
5-4-95:   Check on let/setvar to see if string variable.  Processing to
          handle lists of numbers enclosed as single token disabled in
          this case.  'let ids = "Sample 18, T=200"' works again while
          'setvar a,b,c = "27,38,38"' will also work (more importantly
          'setvar a,b,c,d = &boxcursor'.
4-20-95:  1 pixel error on left side of all symbols fixed (now slightly bigger)
4-20-95:  Double release of memory in vsort.c fixed.
4-20-95:  More problems with the Postscript driver.  Ancient fix to bad
          "include" programs like dvi2ps required a redefinition of 'def'.
          This fix now breaks Postscript Level II.  Decision to tolerate
          only slight bad insertion programs on Level I printers, while
          never breaking Level II printers.  Might say tough, but let us
          know if something else breaks.  Symptoms will be a DictFull error
          on Level I postscript printers.
4-10-95:  Fixed last point being lost in average/squish/compress operations.
4-5-95:   Fixed so "create y = x -range 1 0" (inverted range) works.
3-31-95:  Put in #defines for fgetpos/fsetpos for GNU_EXTENSIONS for
          really brain-dead old versions of GNU C.  Don't think the rest
          of the code has any chance of working with such non-ANSI compilers,
          but at least it will now compile without fatal errors.
3-18-95:  Fixed simple 5-pt smoothing in GENPLOT so less edge effects.
3-18-95:  Fixed so "setv a,b,c,d = &boxcursor" works.
3-18-95:  Fixed problem with \degree, and with many upper characters
          expressed as \xx strings (\thorn, \eta, etc.)
3-17-95:  Fixed problem with negative angle text in PM driver looking like
          italics.  Problem stuffing signed (negative) into unsigned.
3-17-95:  Fixed so that read <0)?0:m  failed.  Modifed parsing of fncs exprs.
1-28-95:  Histogram IDS don't automatically use solid lines.
1-28-95:  create -by  would actually modify the by parameter in order
          to exactly fit the range.  New behavior is to use by exactly and
          create all points that lie within the range.
1-18-95:  Menu on PM screen changed to reflect usage of P for print.
1-18-95:  LEX processor would only switch to mathmode on first try of getting
          a token.  Second attempt after prompt switched back to default
          mode.  "define f(x,y) sin(x)*cos(y)" would fail if you entered the
          f(x,y) after the prompt for a function name.
1-18-95:  Title and timestamp fixed with new fonts.
1-12-95:  Try once more to get the prompting right!  Had been that giving
          both -prompt and -default always took the default without ever
          prompting.  Oh well.
1-6-95:   Added #ifdef around use of Bessel functions in gvcalc.c to handle
          non-Bessel enabled compilers.
12-28-94: Minor change that may have fixed "load pearson" problem.  On my
          machine, "load pearson" is successful from any directory.  .mdl
          files loaded in this way should not require an extension, nor
          being in a specific directory.
12-28-94: Changed the binary mode reading of files so that reading from a pipe
          should be successful in all cases.  Binary files no longer need
          to be seekable.  Use of -binary switch in pipe mode is required
          however.
12-28-94: Per MOT request.  Fixed so "setv a,b=7,8 eval (a>b)?a:b" works.
12-16-94: Per request by Pat Smith, the behavior of &GETARG has reverted to
          an earlier mode.  Giving a -DEFAULT action, but no -PROMPT string
          will cause the DEFAULT to be accepted without user input if no
          token exists on the previous command line.  If you have an option
          to a macro requiring only occassional setting, this can reduce the
          amount of typing.  But it also can be dangerous!
12-7-94:  Fixed spelling of asterisk.  Asterix still accepted though since
          it is my preferred spelling (yes - it is from the comics).
12-7-94:  Compiler optimizer optimized out the Pentium error check so all
          machine report as pentiums with bug.  OOPS.  New version should
          eliminate this annoyance!
12-4-94:  Fixed invalid color problem when linewidth enabled on PM screen
          or when rendering to printer, metafile or clipboard.  Sorry
          folks - got a bit too fancy for myself.
12-4-94:  Changed storage of (REAL *) in HCOPY so should work with double
          precision version of the code.  Use memcpy instead of fancy
          typecasts.  Hope compilers are bright!
12-4-94:  static () the routines in user_aix.c so names don't conflict with
          versions in full RUMP package.
12-4-94:  Fixed so histogram with npoint != 1 works properly.
12-4-94:  HP7000/730 has major problems with X include files.  Appears to be
          related to strict ANSI compliance.  For trial test, set
          XCOPT to same as ECOPT plus defining _XOPEN_SOURCE.  Someone will
          hopefully let me know if this solves the HP7000 problem.
12-4-94:  Fixed minor typecast problems in OSF make (hopefully)
          Still some problem remaining in preprocessor handling of indented
          # declarations.
12-4-94:  Prompt in AUTOY now correct.
12-4-94:  172! and above now return largest representable number (not 1.0)
          Also for negative numbers.  However, since ! has precedence over
          unary -, -100! will not fail.
10-18-94: Fixed problem in first points of TRANFORM AVERAGE and TRANSFORM
          COMPRESS due to early setting of new value.  (Thanks to LRD).
10-14-94: Attempted fix of segment fault when using locmin method in 
          nlsfit two times in a row.  NextGuess not being initialized
          soon enough in code.
10-14-94: alignment problem with allocation of variables manually in big
          block and then local suballocation (first identified in NLSFIT).
          Resolve by using malloc() freely without regard for efficiency
          of heap management.  (Reported and problem identified by MJU)
10-9-94:  gamma(x) optimized with double precision version of GENPLOT.  Now
          max fractional error < 2E-10, average rms fractional error 8E-11.
10-9-94:  All code checked to enable compilation with REAL defined as DOUBLE.
          Can be handled by undef'ing the REAL_IS_FLOAT in sys/mytypes.h,
          cleaning, and rebuilding.
10-8-94:  Fixed lngamma in both real and complex mode so that behaves
          properly with negative arguments.  In real mode lngamma(int x) 
          will return exact for x < 20.
9-29-94:  Fix to xdriver.c code from LRD.  Frees driver structure only after
          X-window fully exits in ENDFNC.  Should eliminate spurious X Error
          messages at shut down or device change.
9-18-94:  Rewrote most of dialog box control in XTERM control windows. 
          Should be transparent at user level, but removes spaghetti code
          and increase supportability.  Maybe now the rest can be completed.
9-02-94:  Fixed whole bunch of pstex devices in devixes.xxx to use subdev
          7 for the encapsulated postscript format.  Should add a device
          "epsf_p" and "epsf_l" for generating portrait and landscape
          encapsulated postscript files.
9-02-94:  Due to problems with some code, the %%EOF end of file marker is
          no longer written for EPSF POSTSCRIPT files.  Also, to improve
          exchange capability, the output from POSTSCRIPT is now in binary.
          This will write only a  instead of  pair at end of
          each line.  os22unix filter not needed to print - other programs
          seem to tolerate either line ending without problems.
9-02-94:  Incorporate LRD's code for a better X-window cursor - tried to
          fix the compile errors on AIX, others can be reasonably expected.
8-29-94:  Increased size of "expression" to 1024 chars in define, let, setv
          so DPB's rediculous expressions don't overrun
8-3-94:   Fixed so echo doesn't come on in NLSFIT when using GENPLOT_MACRO
          or any of the other MACRO processes.
8-1-94:   Fixed so &getarg with no arguments does actually prompt and return
          a value.
8-1-94:   Fixed an unreported error in grid - not sure what the error would
          have done, but it is no more.  Something about if axes were set
          to top or right I think.
6-10-94:  Identified and fixed error in calculation of Q(x^2|v) in both
          the linear and NLSFIT routines.  Oh well, only if x^2>v was there
          an error, and I think few people actually understand it anyway.
          (Kudos to D.P. Brunco for identifying the error.)
6-9-94:   FIT NLSFIT modified to give more accurate error estimates.  The
          errors reported now are consistent with the full statistics given
          by the LINEAR fit.  If no error bar estimates on Y are given, then
          NLSFIT estimates the uniform Y sigma needed to make this a
          "reasonable" fit; error bars on parameters returned for this value.
          If error bars are given, then error estimates on parameters are
          correct.  Q(chi^2) reported also with warning if it falls outside
          the range [0.01,0.99].
6-9-94:   FIT LINEAR modified.  CHISQR$ variable is set to the reduced chi^2
          if sigma error estimates are given.  Otherwise, returns full chi^2.
          This is consistent with the behavior of NLSFIT.
6-7-94:   VSORT is now very quiet by default when started under UNIX.  For
          OS/2, it remains verbose.
6-7-94:   Checks to see if disk really exists when ChangeDisk requested.
          Ie., command m: will fail if m: disk is not real.  There are still
          problems with attaching to a CD when a music disk is there, but you
          shouldn't be listening to music while you work anyway.
6-7-94:   The USER -LOAD dll is no longer automatically released on a 
          "genplot reset" command.  "reset -full" will release it.
6-7-94:   ANNOTE now reports coordinates in %.4g format so works better when 
          in "coormode user" mode.
6-7-94:   Length of the I/O channel string in devices.dat is extended so long
          pipe* commands can be handled.  Now 80 characters.
5-26-94:  Several problems related to loading dynamic modules should have
          been resolved. pearson.mdl no longer requires local DDE4MBS.DLL.
5-22-94:  Minor changes to a large number of files to reduce/eliminate the
          number of compiler warnings.  Improves automated testing of the
          compilation process - but should cause minimal operational changes.
5-20-94:  Fix the HCOPY APPEND problem.  Clearly no-one is using these
          features anymore :-)
5-4-94:   Maximum total # of characters on command lines extended to 2048.
          Single function define f(x) = now allows 1024 characters
5-4-94:   Problem with orientations on PM drivers.  Rotation not recognized
          on the fly in plots - behavior changes with redraw (PS)
4-29-94:  Read of a large number of 3D points generates segment violation.
          Bad pointer modification corrected. (RDR)
4-22-94:  HCOPY SAVE error fixed.  Warning message on exit deleted.
4-16-94:  Fully fixed solid symbols on all LaserJet style output.  Also
          fixed very subtle bug in vsort causing system hang.
4-5-94:   PLOT -HISTOGRAM now recognized with full option name
4-5-94:   LT 4 does not disappear from "no-option" PM screen
4-4-94:   Range in FIX_GRID now unconstrained as in DOS version.
3-28-94:  Fixed the failure of 2^2 that crept in from fixes 3-25-94.
3-25-94:  Modified the exception handling from expression evaluator.  Should
          shut up the underflow exceptions and give valid results (0) for
          underflow conditions.   Also turned off denormalize error so can
          work with numbers to 5E-324, though precision lost below 2.225E-308.
3-21-94:  Added GENPLOT as a default RETURN from CONFIG
3-21-94:  Fixed so "eval 3==sqrt(9)" works as expected
3-21-94:  Fixed fix_grid so range can include entire X range
3-9-94:   Fixed black page printing with postscript driver.
2-27-94:  More rigid checking of disk change command.  C: or D: alone is now
          required.  Previously, any path beginning with x: was recognized
          as a disk change - these now generate errors.
2-15-94:  Moderate speedup of some math routines by inline math functions
          instead of function calls.  FFT benefits in particular.
1-27-94:  Enable 3D axes.  They work as far as I can tell including the HCOPY
          redraws.  Only problem is with massive skew, but too bad.  Will not
          be fixed.  Only question is whether X,Y axis should flip "down" for
          tilts beyond +/- 45 degrees, as happens for Z.
1-24-94:  Fixed so "cd /" now works.  "cd \" still fails because the \ is
          read as request for continuation char.  Always use / in UNIX and
          OS/2.
1-24-94:  Fixed labelling of non-linear Y axis with some increment values.
1-19-94:  Maybe fixed strange error that caused first line of some data files
          to be ignored.  fseek(lun, 0L, SEEK_SET) replaced by rewind(lun).
          This will be reported as a compiler bug.
1-19-94:  Eliminated annoying crash of SVOUT on some operations.  Easiest to
          demo was:  "axis hc dev null shrink 1.6".  No longer crashes.
1-19-94:  Solved crash that occured when bad device driver specified (LRD)
1-5-94:   Fixed bug so 3D fitting allows "equ f(x,y)" without added quotes
1-4-94:   Retrieve of a 3D structure now switches GENPLOT into 3D mode
1-4-94:   Cull'ing of a 3D structure now works properly, though maybe not
          exactly as expected for average, compress etc.
9-23-93:  Fixed POSTDRV driver in sub-device 8 mode.  Negative bounding box
          values overwriting the %% comment of PAGES comment.
5-18-93:  Timestamp indicated 12 hour with no AM/PM.  Default is now 24 hour
          but can be changed using "let $TSTAMP:FORMAT = '%X on %x'"
1-15-93:  Palette choice in PM graph screen now saved in profile.  Device
          options in DEVICES.DAT has no effect now.
6-10-93:  Changed output from ANNOTE cursor report so X and Y don't ever
          run together without a space in between.
6-17-93:  Made sample draw (2draw, 1draw) close straight.  Still needs
          arbitrary angles.

===========================================================================
FIXED PROBLEMS:
===========================================================================
Report: 9-1-92          From:   MOT                     Prty:   1
Error:  Lines of graph are lost if DEV PM and PLOT commands on same line
Source: The threads for PM drawing appear not to be fully initialized before
        the first drawing commands come through from the main program.  Will
        need another semaphore to indicate when actual drawing is available.
FIX:    3/1/94 - think this is resolved.

Report: old
Error:  Multiple graphs to postscript printer cause subsequent pages to be
        printed shrunk and rotated.
Fix:    Added document level save and restore of state around entire graph
        component.  No longer requires printer reset by spooling software.

Report: old
Error:  If you kill a command line (^K), you can't get it back with yank (^Y)
Fix:    Can do now (I think)

Report: old
Error:  If the HCOPY vector save file filled up or generated an error, the
        file would remain open and could not be deleted until GENPLOT ended.
Fix:    Code now attempts to robustly close and delete the HCOPY file. 
        Message indicates fully the SAVE is now inactive.

Report: old
Error:  Linewidth control did not work on PM screen
Fix:    Finally figured out how to do it.  Works well, and should work with
        all hardcopy devices as well.  On screen, it slows things down a fair
        bit so must be enabled on options before run.  There is another
        control bit that is supposed to use thick lines when drawing.  Not
        clear if this works, but it slows things down so must be doing
        something.  See new options menu on PM graph screen.

Report: 11-18-93        From:   linux users             Prty:   3
Error:  CONWaitChr fails causing help to be non-functional.  No proper
        implementation of setitimer() in linux GCC compiler.
Fix:    Code for CONWaitChr has been completely rewritten to utilize
        only POSIX terminal control operations, CANON and VTIME/VMIN
        controls.  WARNING: THIS MAY BREAK OTHER NON-COMPLIANT COMPILERS.
        SEE THE TEST CODE IN NEW DIRECTORY UNIX_TESTS.

Report: 07-xx-93        From:   LRD                     Prty:   1
Error:  X display would crash if you tried to plot immediately after open
Fix:    Problem traced to XdrawPoints() and XDrawSegments() being called
        with 0 valid pts.  Do own check to avoid crash (Fix 8/3/93 LRD)

Report: 07-28-93        From:   LRD                     Prty:   3
Error:  FIX_GRID fails if you specify the range
Fix:    Bad programming - shoot the author.

Report: 06-29-93        From:   KB                      Prty:   4
Error:  Inequality comparisons backward in mathmode complex mode.
Fix:    Bad programming - shoot the author.

Report: 06-13-93        From:   RDR                     Prty:   3
Error:  -errx, etc. error bars no longer work.
Fix:    Determine length of function command buffer before resetting the
        starting point.  After resetting, length appeared to be 0.

Report: 10-29-92        From:   Thompson                Prty:   2
Error:  If ^C typed while waiting for input on command line, program will halt
        with SYS3171 error.  This is compiler bug in the getch() routine and
        there is no indication if/when it will be fixed.  The problem will
        disappear anyway with the introduction of the PM window interface.
Fix:    New compiler fix has eliminated problem, mostly.  After ^C, break is
        not recognized until the next character.  Solves much of the problem,
        though not as elegant as the PM window will be.

Report: 06-01-93        From:   MOT
Error:  xeq now2 fails (as does any name starting with NOW for macros)
Fix:    Running the internal created macro only works with "xeq n", "xeq now" or "xeq /".

Report: 06-01-93        From:   MOT
Error:  UNIX version asks that you end comments and macros with a ^Z
Fix:    Modified prompts so ^Z is used for OS/2, but ^D is used for UNIX environments.

Report: 03-01-93        From:   LRD                     Prty:   2
Error:  On LINUX, fails to byte reverse properly on binary data file write
Fix:    Original code byte reversed only on OS2 operating systems.  Code now
        changed so looks are architecture type.  Currently i486 is recognized
        as byte reversed format.  All others are assumed to have normal byte
        order (high byte to low byte on disk).  This will affect ability to
        read older data files written on LINUX.
WARNING: All machine architecture names have changed to be consistent with
         other software.  In particular, names are now nominally lowercase
         with many more defined.

Report: 05-10-93        From:   MOT                     Prty:   999
Error:  sqrt(-1) returns -j
Fix:    Default branch cut was taken along the negative real axis, so
        sqrt(-1) could be + or -j.  Fixed by making the branch cut a
        selectable paramter.  Added "MATHMODE BRANCH [positive | negative]"
        as well as "MATHMODE STATUS".

Report: 05-07-93        From:   MJH                     Prty:   2
Error:  HPGL driver no longer does any pen colors except pen 1
Fix:    min/max exchanged in limit instruction

Report: 05-06-93        From:   DPB                     Prty:   1
Error:  In function definitions, special characters not allowed in dummy
        variables.  define f(m_m) = fails.
Fix:    Add special characters to allowed set [0-9][a-z][:$_]

Report: 05-06-93        From:   KB                      Prty:   2
Error:  define f(x) = {x} fails to recognize the replacement variable
Fix:    Correct handling of {}[] as well as () in define expressions

Report: 03-17-93        From:   RDR                     Prty:   2
Error:  Following macro would fail
           autoerase off erase
           size 7 7 axis hcopy dev /
Fix:    Don't leave extraneous code laying about

Report: 03-17-93        From:   RDR                     Prty:   2
Error:  Portrait mode on Postscript driver has problems
Fix:    Change handling of size so reports proper limits to GENPLOT

Report: 02-28-93        From:   BL                      Prty:   3
Error:  @std(y) returns wrong value
Fix:    Learn to code correctly, twit!

Report: 02-17-93        From:   BL                      Prty:   2
Error:  GENPLOT crashes drawing string '{abc}{def}'
Fix:    Well, let's pass addresses if I want to get information back from
        a subroutine.  Fixed it, stupid!

Report: 02-17-93        From:   AV's partner            Prty:   2
Error:  Plots to POSTSCRIPT fail with new driver
Fix:    Command stack was being flushed before the page was "drawn". 
        Exchanged the order of the those commands and all is okay.

Report: 02-5-93         From:   JSC                     Prty:   3
Error:  Functions calling functions with args fail.  g(a,f(b)) for example.
Fix:    Critical failure in stack maintenance.  Exhibited as wierd math
           define f(x) = x define g(x,y) = x+y
           eval g(f(10),0) works, but eval g(0,f(10)) fails.
        Parsing modified to keep better track of internal evaluation stacks

Report: 02-4-93         From:   DPB                     Prty:   1
Error:  Transforms SQUISH and COMPRESS are reversed in sense from DOS version
Fix:    Okay, now switched.  Wish I could remember which did what without
        the help.

Report: 02-1-93         From:   LRT                     Prty:   0
Error:  GENPLOT now starts with curve of 2048 points, all undefined
Fix:    This arose when RDR requested curves be "allocated" with size set
        to maximum.  Special case on initialization now resets number of
        points to 0.

Report: 01-25-92        From:   LRD/DPB                 Prty:   0
Error:  The format (E12.6) doesn't work in &ENCODE - compatibility problem
Fix:    Okay, now it does.

Report: 01-25-92        From:   LRD/DPB                 Prty:   0
Error:  \circ in symbol table is inconsistent with TeX format
Fix:    \circ, \triangle and \star are now defined and similar to TeX
        \degree replaces usage of \circ for degree symbol (\degree C).
        Also, added the thin space command \, as in TeX.

Report: 01-15-92        From:   RDR                     Prty:   0
Error:  The prompt in an &GETARG structure doesn't allow %xx% replacements
Fix:    Okay, now it does.

Report: 01-15-92        From:   TJ                      Prty:   1
Error:  genplot_.ini is created in current directory rather than on path
Fix:    Has been fixed.  Failed to use os2.ini variable in code other than
        main routine.  It should open the first genplot_.ini that it finds
        in the path when it starts.  This should be in
        /usr/genplot/os2/genplot_.ini.  Saving the profile should overwrite
        the one already open.

Report: 1-1-92          From:   MJU                     Prty:   0
Error:  Postscript output is not available in encapsulated format.
Fix:    Incorporated much of DCS level 3.0 code from MJU as new postscript
        driver.  Devices 1-5 are normal postscript output, and 6-10 are
        encapsulated output.  Output conforms to DCS level 3.

Report: 9-8-92          From:   JSC                     Prty:   1
Error:  OS/2 Postscript driver places graph slightly different than DOS
Fix:    Entire postscript code has been rewritten (see above). If this is 
        still a problem, it is now a new one.  Old report is now defunct.

Report: 1-15-93         From:   LRT                     Prty:   1
Error:  CURSOR -TRACK has problems with pressing arrow keys too fast
Fix:    Any key pressed within 200 msec of a return from any cursor function
        will now be ignored.

Report: 1-13-93         From:   LRD, DS                 Prty:   1
Error:  Minor corrections to UNIX makefiles.
Fix:    (1) Made GNU_EXTENSIONS consistent in all files (OOPS)
        (2) Added (char *) casts to read() and write() calls
        (3) Fixed spelling errors in abusive error messages

Report:                 From:   Annil                   Prty:   2
Error:  Help is non-functional on UNIX based machines.
Fix:    After several tries, revised help file to include indexes based on
        both big and little endien architectures.  Should automatically
        detect and handle either type machine in UNIX environment.

Report: 1-13-93         From:   JK                      Prty:   1
Error:  Create form y=f(x) no longer works. y= f(x), y =f(x) and y = f(x) okay
Fix:    How many different syntax choices can there be.  I think they are
        all enumerated now.

Report: 1-12-93         From:   JK                      Prty:   0
Error:  Under some circumstances, ls will put machine into lost state with
        continuous SEGMENTATION FAULTS.
Fix:    Very subtle bug - only occured if all files were of xxxx.aaa format
        exactly 8 character long.  Memory overwrite fixed.

Report: 1-10-93         From:   LRT                     Prty:   1
Error:  HCOPY UNDO will undo the effect of an IDENTIFY -PLACE if executed
        shortly after being place is set.
Fix:    Moved TIMESTAMP and IDENTIFY control variables out of HCOPY save
        structure.  Are now static relative to HCOPY, but dynamic with respect
        to multiple plot windows.

Report: 10-29-92        From:   Thompson                Prty:   2
Error:  (1) Axis scales change each time you plot.  Set range of -7,7.
            First time gives range -8,8 but next time stabilizes to -10 10.
        (2) Top axis often out of sync with bottom axis - see 1 above.
        (3) XTOP off does not completely track bottom.  AXCTRL will only
            affect the dx/dx2 spacings of the specified axis.
        (4) Grid fails to use the DX spacings set by AXCTRL.
        (5) Y axis labels never shrink in size, even if DX made very small
Fix:    The autoscaling routine now iterates until it finds a stable solution.
        This should solve problems 1 and 2.  -7,7 will always go to -10,10.
        0 to 14 will always go 0 to 14.   AXIS now checks first tops own
        axctrl() settings, then uses BOTTOM's if in tracking mode.  Made
        grid look at axctrl settings also.  Y labels now rescale properly if
        too many included.

Report: early           From:   MOT/RDR                 Prty:   1
Error:  Keyboard keystrokes do not work with cursor on PM device.
Fix:    Key strokes are now trapped properly for moving the cursor in normal
        and box cursor modes.  It should operate again like the DOS cursor
        except that the "f" key will not free the cursor when in tracking
        mode.  You must use the mouse to go into find mode in cursor -track.

Report: 1-04-93         From:   RDR                     Prty:   1
Error:  DOS allocates curves assuming filled with data, OS2 as empty
Fix:    Changed so ALLOCATE C1 will return with arrays zero and C1:NPT
        set to C1:NPTMAX.  All arrays and curves now calloc() instead of
        malloc() so will be initialized to zero (assuming IEEE floating)

Report: 12-20-92        From:   RDR                     Prty:   2
Error:  Crash in SPLINE -SMOOTH for large npt, small 
Fix:    Underflow condition on matrix inversion.  Promoted all args to
        double - will avoid crash for reasonable conditions.

Report: 12-16-92        From:   SRussell                Prty:   3
Error:  Complex power function fails (z**2)
Fix:    OOPS - that's what beta tests are for.  z*z was okay, z**2 bad.

Report: 11-xx-92        From:   Joo                     Prty:   3
Error:  Spline fails to extend properly below lowest X value
Fix:    This was fixed with new code earlier, but I left the offending code
        in the file immediately after the fix.  Sorry.  Spline should be okay
        now.

Report: 12-10-92        From:   Shirish                 Prty:   1
Error:  The ^D not working for command line editing is a pain in the butt
Fix:    ^D is the UNIX end of file specifier.  Code modified so that EOF
        is only recognized as the first character of an otherwise blank
        input line.

Report: 10-29-92        From:   TJ                      Prty:   1
Errors: The ^C can only be used once - second time aborts program
Fix:    Again, stupid IBM C compiler - requires resetting signal() after
        every use.  Done.  However, still have problems with ^C crashing if
        waiting for text input.

Report: 11-01-92        From:   RDR                    Prty:   2
Errors: FILE*DEFAULT opens a file "default" instead of temporary filename
Fix:    11-05-92
        OOPS! Forget that little detail.

Report: 11-01-92        From:   RDR                    Prty:   2
Errors: Generating more than one plot to LaserJet crashes GENPLOT
Fix:    11-05-92
        Error attempting to write to a closed file handle.  vsort.run now
        remains active until the handle is closed by GENPLOT exiting.  Error
        in file deletion appears to be C bug - added sleep(1) to wait after
        closing .vec file before attempting to unlink() it.

Report: 10-31-92        From:   RDR                     Prty:   1
Error:  Complex commands missing the ARG(z) function.
Fix:    11-05-92
        Added it.  Value of 0 return 0 as arg, otherwise ln(z/abs(z))

Report: 10-23-92        From:   Thompson                Prty:   3
Errors: Apparent stack overflow in complex math trashing file reads
Fix:    10-24-92
        Allocated correct type variable in gv_eval_complex_expression()

Report: 10-20-92        From:   Thompson                Prty:   3
Errors: min(x,y), max(x,y) fail in complex mode
        setvar of an already complex variable gives error, but works correct
Fix:    10-20-92
        Small modifications to the complex code.  MIN/MAX now check only the
        real component.  If real components are equal, then checks imag.
        RND(x) now returns only real component, imag = 0.  Allows ndtri(rnd(x)).

Report: long ago        From:   Everyone                Prty:   2
Errors: Unable to use cursor from device XTERM on X-window terminal emulator
Fix:    10-4-92
        Screwup in the IO_GetInput routine.  Assumed all the characters would
        be read when in fact only a single character was read.  Also fixed so
        left mouse button returns '0' and right returns '1', corresponding to
        expected return of a  for repeat cursor commands.

Report: while back      From:   MOT                     Prty:   2
Error:  Macro's cannot be executed until a base directory is specified.
Fix:    25-Sep-92
        Fixed.  Also added aliases SAVEMacro and BASEMacro to SAVE_Macor and
        BASE_Macro commands.

Report: while back      From:   DWS                     Prty:   2
Errors: Impossible to exit from entering text for MACRO and WRITE -BINARY
Fix:    10-4-92
        Failed to define an "end-of-file" character in the keyboard routine.
        ^D now works on UNIX systems just like ^Z on the OS/2 systems.  Screen
        will show ^Z always, but must use ^D on UNIX systems.

Report: various         From:   various                 Prty:   2
Errors: numerous oddities (axis labels to 1.09 for instance)
Fix:    9-16-92
        There were strange occurances with some numbers near one.  Exhibited
        as things like 1.09 on graph where 1.0 was expected.  COMPILER BUG!
        Installed compiler patches 9/16 and these bugs disappeared!

Report: 9-8-92          From:   DPB                     Prty:   3
Error:  cotd(x) returned sind() instead of 1/tand(x)
Fix:    9-8-92                  
        Programmer failure (not enough sleep obviously)

Report: 9-8-92          From:   RDR                     Prty:   2
Error:  axis -grid failed for "reg left 1 0 reg bot 1 0"
        axctrl left -grid failed for reg left 1 0
Fix:    9-8-92
        Programmer failure (not enough sleep obviously)

Report: 9-22-92         From:   JSC                     Prty:   3
Error:  The gamma(x) function does not return gamma.  
Fix:    9-23-92
        AARGH!  Believe it or not, IBM CSET/2 provides a gamma() intrinsic 
        function  that returns ln(|gamma(x)|) for x>=0.  So much for functions 
        doing  what their name implies.  Since they also disallow - arguments, 
        it is back to my routines.  Not sure slower or faster, but fixed.

Report: 9-22-92         From:   JSC                     Prty:   3
Error:  setv t = 2**-127 fails,  t=2**-126 succeeds
Fix:    9-23-92
        Problem is storing a number <1E-38 into REAL*4 space.  The compiler is
        supposed to simply store as 0, but instead gives math exception.  Now
        testing for < FLT_MIN and storing 0 if it is too small, no warning!

Report: 9-22-92         From:   JSC                     Prty:   2
Error:  "sgraph list" failed miserably with exception fault
Fix:    9-23-92
        Caused by passing an unitialized variable.  ENHANCEMENT:  Added the ? 
        as a response to sgraph to do list.  Commands "sgraph ?", "sgraph /" 
        and sgraph "list" are all synonymous now.

Report: 8-1-92          From:   MOT                     Prty:   1
Error:  "macro" fails unless BASE_Macro set, no error accompanies failure
Fix:    9-24-92
        Empty base directory name was not expected.  Changed behavior so that
        without Base_Macro set, default is to store new macro file in the
        TEMP directory (pointed to by TEMP or TMP environment variables).
        Added aliases BASEMacro for BASE_MACRO and SAVEMacro for SAVE_MACRO.

Report: 9-23-92         From:   MS                      Prty:   1
Error:  Aliased commands are not recognized inside NLSFIT
Fix:    9-24-92
        Added the check for an alias to NLSFIT, ANNOTE and SAMPLE
        subprocessors.  Essentially, anywhere the command processor is sticky,
        ie. hitting just a return leaves you in the processor, an aliased
        command will be recognized.  Note that ALIASES are only expanded if
        they occur at the command level.  ie., ANNOTE GO will expand GO into
        it's alias, while ANNOTE LABEL GO will not recognize GO as an alias.

Report: 10-01-92        From:   RDR                     Prty:   0
Errors: Request sign(x) return -1 for x<0, 0 for x==0, and 1 for x>0
Fix:    In absence of any pressing need for the FORTRAN standard, as it is
        currently configured, change has been made.  Let me know if this is
        a problem.

Report: 06-24-92        From:   MOT                     Prty:   2
Error:  Default cursor position failed to be recognized in ANNOTE
Fix:    Removed use of -999.9 as default value since it is not exact in
        IEEE FP representation.  Switched to FLMAGIC parameter (currently
        1824724) for the default value.

Report: 06-24-92        From:   DPB                     Prty:   2
Error:  PMDRIV will fail printing plot if device closed shortly after request
Fix:    Modified code so PMDRIV will no terminate until after print is
        complete.  Still may be a problem if you request multiple plots from
        the screen device.