From octave-graphics-request at bevo dot che dot wisc dot edu  Thu Feb 10 03:12:05 2000
Subject: Adding output for other plotting programs
From: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
To: Peter Hopfgartner <hopfgartner at rolmail dot net>
Cc: octave-graphics at bevo dot che dot wisc dot edu
Date: Thu, 10 Feb 2000 03:11:53 -0600 (CST)

On 27-Dec-1999, Peter Hopfgartner <hopfgartner at rolmail dot net> wrote:

| I've posted recently some scripts to output results with superficie
| (plot3, quiver3, surf) and played a bit with an output with the
| GNU plotutils.
| 
| Anyway, I was wondering, if there is any kinf of infrastructure in
| Octave 2.1.x for multiple packages.
| 
| In particular, how should these kind of new interfaces be added.
| 
| Should the plot/mesh etc command read an environment variable to chose
| which package to use? Or should they simply have sightlly different
| names?
| 
| How should I access the informations for titles, axis labels etc?

These are all good questions, and unfortunately I don't have have good
answers for them.

| Is this work welcomed?

Yes.

I think the plotting interfaces will eventually have to be rewritten
almost completely.  The current interface to gnuplot is really
starting to bug me.  The following outlines one reason why (I sent it
to the gnuplot maintainer).

  I finally started trying to have Octave send data to gnuplot without
  creating temporary files and ran into some problems.

  Currently, for each pair of x-y data in a command like

    plot (x1, y1, x2, y2, x3, y3, ..., xn, yn)

  Octave creates a temporary file and then generates a command like

    plot "/tmp/fileXXX"

  for the first line and

    replot "/tmp/fileYYY"

  for the remaining lines.

  This works fine for data stored in files, but it doesn't work if the
  data are sent to gnuplot along with the commands.  In other words,

    plot "/tmp/fileXXX"
    replot "/tmp/fileYYY"

  works, but

    plot '-'
    1 2
    ...
    e
    replot '-'
    3 4
    ...
    e

  does not (it seems that gnuplot expects to see the data for the first
  line again).

  For this particular case, Octave could be fixed to save all the data
  in temporary files and then generate a single plot command that looks
  like this:

    plot "/tmp/fileXXX", "/tmp/fileYYY"

  If that were all, then I would go ahead and try to make these changes,
  but there is one more twist.  Octave has a feature (patterned after a
  similar feature in Matlab) that allows one to add lines to an existing
  plot by using

    plot (x1, y1)
    hold on
    plot (x2, y2)

  In this case, the first plot must be displayed, then the second line
  added to it.  Currently, Octave's method of using temporary files does
  this correctly.  The first plot command generates a temporary file and
  a

    plot "/tmp/fileXXX"

  command for gnuplot.  Then the second plot command generates another
  temporary file and a

    replot "/tmp/fileYYY"

  for gnuplot (even though it is the first curve for the current call to
  the plot function, `replot' is used instead of `plot' because the
  `hold' state is `on').  This works great.  Gnuplot remembers the name
  of the previous file and both lines are displayed on the plot.  But if
  I send the data along with the plot commands, because gnuplot wants to
  see all the data and I only have one set (x2, y2).

  The fix is not impossible, but it is a little messy.  Octave would
  have to hang on to the last set of plot data just in case a new plot
  might be made with the `hold' state `on'.  If there is no other way, I
  can probably arrange for this to work in Octave, but since gnuplot
  already remembers the names of files that it has previously plotted,
  would it be possible, when the data are coming from the command
  stream, to have it remember the data that it last used as well?

  What do you think?

  Thanks,

  jwe


So, I'm ready to start thinking about how to improve things so that
other plotting packages can be easily supported within Octave.  I
don't care about supporting the gplot/gsplot interface with other
plotting packages though, so people who are using that will have to be
stuck with gnuplot.  Anything else would be too hard, I think.  But
people who use the plot() commands (or some other better set of
plotting functions) should not be forced to use any particular
plotting package.

The design needs to relatively clean.  We can't have every plotting
package implementing every detail of plotting.  Lots of code needs to
be shared, otherwise I think it is too hard to incorporate new
packages.

Any ideas?

jwe