Category : C Source Code
Archive   : GPLTS202.ZIP
Filename : README2.PAT

README2.patch - some notes about gnuplot 2.0 patchlevel 2.

The major additions in this patch are parametric functions,
X11 Motif support, new bit mapped graphics routines, and new terminal
drivers vttek (VT like tektronix emulators), hpljii (HP LaserJet II),
kyo (Kyocera Laser Printer), SCO CGI.
For a more complete list of the changes, see the MODIFICATIONS
section below.


APPLYING PATCH 2
----------------

To apply this patch copy the files README2.patch, patch2a, patch2b,
patch2c, patch2d, patch2e and patch2f to the gnuplot directory.
Change directory to the gnuplot directory, and then apply the patch with
the following commands:

rm term/font5x7.trm
rm term/eps60.trm
rm term/hpljet.trm
rm term/epson.trm
mv README.patch1 README1.patch
patch -p < patch2a
patch -p < patch2b
patch -p < patch2c
patch -p < patch2d
patch -p < patch2e
patch -p < patch2f

Copy the appropriate makefile.* to Makefile, READ THE Makefile
and make any necessary changes. Then recompile!


MAIL ADDRESSES
--------------

The mailing list info-gnuplot exists as a forum for discussions about
gnuplot. About the only gnuplot topic not supported by this list is
bug reports. They are handled by a different mail alias, bug-gnuplot.

To be removed or add someone to the mailing list send mail to:

[email protected]


To report bugs send mail to:

[email protected]


To ask a question or discuss gnuplot send mail to:

[email protected]


pixar is a UUCP site. sun.com is a UUCP and Internet site.
In the above addresses, sun.com can be replaced by ucbvax.berkeley.edu.


PARAMETRIC IMPLEMENTATION
-------------------------

At the request of Bill Schulz, one of our illustrious math profs, I
was "encouraged" to build parametric plotting capability into gnuplot.
He not only explained how much more useful parametric plotting was
compared to the simpler case of function plotting, but he suggested
the syntax for parametric plots as simply x,y pairs of functions on
the plot line. (He also suggested that the changes must be very easy
to do, but I forgive him for that bit of chicanery.)

Thus armed, I was surprised at the number of other issues I ran into.
For instance, in addition to inventing a `set parametric' command I
took a hard look at xrange and yrange and decided I also
needed a trange when in parametric mode. This, happily, led
to what I believe is the right behavior in all cases I have tested
for parametric equations. Indeed, I feel the implementation is
better than I originally hoped for as you can, in parametric mode,
control the t range, x range and y range independently. This gives
you quite a bit of control over the function being plotted.

If the parametric commentary seems to be too verbose I'm willing to shorten
it on command. (I normally don't write like that--geesh!) I'm happy to
leave it alone, however, if no one finds it offensive.
Of course, to be complete I needed an autoscale_t variable (such already
existed for x and y). The meaning of this variable, however, was a bit
weak. Internally it is necessary, but from the user's standpoint,
it may not make much sense to have a `set autoscale t' command. All that
gains the user is a chance to catch an empty trange and expand it if
necessary. How an empty trange comes into existence is a mystery
to me. (Empty x ranges can occur, I guess, due to a data file or
being in polar mode.)

The trange, by the way, has an arbitrary default of [-5:5] (whereas
the xrange default is [-10:10]). This was a range suggested by Bill
Schulz when I pinned him down on what would be "reasonable" one day.
If you can think of any reason for any other range please let me know.
The current range has the habit, by the way, of plotting over cyclic
functions defined on [0:2*pi] such that dashed lines don't look good.
There seems to be no real "solution" other than to avoid the default
range where it doesn't make any sense.

Plotting, as described in my additions to gnuplot.doc, in parametric
mode requires pairs of functions. Consider the following:

plot [-3:3] [0:20] [-10:20] sin(t),t**2,'data.file',t,t**2

which has a trange of [-3:3] an xrange of [0:20] and a yrange of [-10:20].
The plot consists of 3 curves: sin(t),t**2; the 'data.file' points;
and the plot of t,t**2. The last, of course, could be done in
non-parametric mode as simply x**2. In parametric mode, you have to
be finished defining a parametric pair (an x-function and a y-function)
before you can put in a `with' or `title' modifier.

Ranges can be set on the plot command or with the `set [xyt]range'
command. As an added benefit, when in parametric mode, the xrange
can be autoscaled. This is true because, like the yrange, the xrange
is now a computed entity.

Finally, I decided that the data file, as ordered (x,y) pairs, could affect
the xrange but not the trange. Hence plotting the following:

plot sin(t),cos(t),'data.file'

Where 'data.file' only has points in the positive x range will cause
the plot of sin(t),cos(t) to take on x values only in the positive
x range. Another option was to make 'data.file' x values affect the
working t range of the parametric function. This did not seem like
a good idea, however, and some effort was made to avoid this affect.
The best plan might have been to ignore the 'data.file' ranges
all together when in parametric mode. I hope someone will think
on this issues and voice an opinion at some point in the future.


UNFINISHED ISSUES

There are some unfinished issues that I ran across in doing the parametric
implementation. These are not so much parametric related issues as
they are general gnuplot issues that the parametric work brought to the
forefront.

First, what is the use of global xmin and xmax changes when a
data file is plotted? In other words, why would xmin and xmax
change for the whole session when the data file was just mentioned
once on a command line. It seems that this global behavior is
unwarranted. The routine in question, get_data() simply checks
against the global xmin,xmax rather than the local x_min, x_max
that eval_plots() will eventually pass on to do_plot(). The change
wouldn't be hard to do, and I'm hoping someone will decide it is
a good idea for a later bug patch.

Also, the default style for each individual plot depends upon the
line_type and point_type of the plot. Gnuplot currently increments
the line_type (and possibly point_type) for each plot that it is
plotting regardless of the type of plot in question. For instance,
plot x,'data.file',x**2 will generate x with line_type of 1 and
x**2 with line_type of 3 while plot x,x**2,'data.file' will use a
line_type of 1 and 2 respectively. This also seemed expedient, but
not quite right in terms of what would be nicest behavior. It seems
to me that line_type's are scarce resources and should be used
(incremented) only as needed. In other words, since the 'data.file'
plot isn't using a line_type at all, the second line_type should be
saved for the x**2 plot, and the two plots would look the same
instead of being different as they are now.

Of course, neither of these issues are critical. We've been living with
them comfortably for the whole life of at least the latest gnuplot 2.0
release. It seems to me, however, that both these problems are minor
oversights and could easily be corrected in the near future. Rather
than do that here, as part of the parametric work, I decided that would
be a change to the previous non-parametric behavior and should not be
"snuck in" without a bit of debate.

John Campbell [email protected] or [email protected]


SCO CGI DRIVERS
---------------
copyright 1990 Ronald Florence

The 'cgi' driver added to Gnuplot 2.0 will work with the SCO CGI drivers.
To use the CGI drivers, you will need to have the CGIPATH and CGIDISP
and/or CGIPRNT environmental variables declared. The CGI output mode
can be controlled by setting an ASPECT environmental variable. The
default is to use full-screen/full-page.

Two CGI terminal options are provided. If you have either CGIDISP or
CGIPRNT declared, Gnuplot 2.0 will start with terminal set to "cgi".
If both devices are declared, you can "set terminal hcgi" to select
the CGIPRNT (hardcopy) device instead of the default CGIDISP device.
This is useful when you are proofing a graph on the display before
sending output to a printer or plotter.

The driver has been tested with the SCO vga, ega, cga, hercules, and
laserjet drivers. If you are using a monochrome VGA display you may
need to include "-DVGA_MONO" in your Makefile to work-around the
broken SCO monochrome VGA display drivers. The CGI driver includes
a built-in pause-for-any-key feature which makes the Gnuplot
"pause -1 prompt" command redundant.

Permission is hereby granted for unlimited non-commercial use of this
code, on condition that the copyright notices are left intact and any
modifications to the source code are noted as such. No warranty of
any kind is implied or granted for this material.

Please send suggestions and comments to [email protected].


MODIFICATIONS IN PATCH 2.
-------------------------

Tony McGrath
Fixed replot line length calculation (command.c).

John Engel
vttek terminal driver (term/tek.trm).
vms additions for tektronix drivers (term.c).
hpljii driver - now uses the same bitmap routines as the epson driver
and will work on a PC (term/hpljii.trm).

Russell Lang
Created term.c function 'void reopen_binary()' for PC (term.c).
Removed HPLJET driver.
Added an error check for bad xrange when plotting functions -
this came up when an attempt is made to plot an non-existent
data file with x autorange, followed on the next command line
by an attempt to plot a function (command.c).
Added new bit mapped graphics (bitmap.h, bitmap.c, term.c,
epson.trm, hpljii.trm).
Removed PROPRINTER driver - use epson_lx800 instead.
nec_cp6 monochrome drivers also work with epson lq-800.

Rik Harris
Fixed 'set dummy' command (command.c).
Previously crashed if dummy variable missing.

Dave Kotz
added x11 autosense code (term.c).
empty gnuplot.doc lines now produce empty lines in .ms
or .tex files. (doc2ms.c, doc2tex.c)
check for zero argument in call to x**(-n). (internal.c)

James Dugal
Fixed messy octal constants (epson.trm).

Jeff Kellem
Fixed ceil and floor to work as per documentation - for complex argument,
return integer ceil/floor of real part (standard.c).
Changed gnuplot_x11.c to use default screen (gnuplot_x11.c).

Anthony Thyssen
(Ideas by Anthony, code by Russell Lang)
Changed 'plot with impulses' to draw impulses to origin (graphics.c).
Added 'plot "datafile" using y|xy|yx "scanf string"' (command.c).

Greg Montgomery
Turbo C BGI drivers now use far code, to avoid segment _TEXT
overflow with TC++ (makefile.tc, linkopt.tc, plot.c).

Sebastian Kremer
Kyocera laser printer driver (term/kyo.trm).

John Eaton
set [no][xy]zeroaxis (graphics.c, setshow.c, setshow.h).

John Campbell
Parametric plots (command.c, misc.c, setshow.c, setshow.h).

Ronald Florence
SCO CGI driver (term.c, term/cgi.trm).

Chip Rosenthal
Fixes a security problem with ".gnuplot" files (plot.c).
Add pipe option for output file on Unix. Can now say
'set output "|more" '.

Ed Kubatis
Ignore interrupts in gnuplot_x11.c (gnuplot_x11.c).
Added X11 Motif support (gnuplot_x11.c).
Demo files for 'plot using'.
Added ifdef GETCWD for HP-UX (command.c).
Fixed RS/6000 AIX 3.1 build problem (misc.c).

Ronald Hartranft
Base work on new bit mapped graphics routines.

Alex Woo
Changed -DGAMMA to -DGAMMA=gamma so that -DGAMMA=lgamma
can be used on the NeXT (standard.c, makefiles).

End of README2.patch