AW: Very slow ploting with gnu plot (windows)

[Schirmacher, Rolf]

Hello,

for plotting I see the following timing (running octave directly from
the terminal window as by standard installation)
Gnuplot used here is the one delivered with the octave installation
(gnuplot 4.4 pl 4), there is no other gnuplot.exe on my system at the
moment.

GNU Octave, version 3.6.2

Octave was configured for "i686-pc-mingw32".


"Small" plot test:

tic; plot(sin(0:0.01:10)); toc

Plot times are about 0.1 seconds for a new figure window and 0.06
seconds to replot into an existing one with some scattering and more or
less independent of the graphics toolkit.  Plot opens "immediately"
(subjectively) with fltk and qt, but take a noticeable time (1-2s ?) to
build up OR REFRESH with gnuplot.


"Complex" plot test:

plot_data = rand(100);
tic; plot(plot_data); toc

Gnuplot: first plot: elapsed time 3 s, octave prompt re-appears after
additionally 5 s, plot shows up after additionally 1 s. 
                   Replot: elapsed time 3 s,  octave prompt re-appears
after additionally 10 s, plot refreshes after additionally 1 s. 
fltk:           first plot: elapsed time 3s, octave prompt re-appears
immediately, plot shows up immediately.
                   replot: as first plot.
qt:             same as fltk.


"Very complex" plot test:

plot_data = rand(100);
tic; plot(plot_data); toc

Gnuplot: Elapsed time 30s, prompt re-appears after additional 40
seconds, plot shows up additional 60 s later.
fltk:           Elapsed time 30s, prompt re-appears immediately, plot
shows up immediately
qt:             Elapsed time 30s, prompt re-appears immediately, plot
shows up immediately


For me, it looks like a communication issue in between core octave and
gnuplot including the octave side (slow response of the octave prompt
coming back), plus perhaps an extra gnuplot issue with display speed (or
the second half of the communication? But display speed is involved: Try
to rescale the plot window for the "very complex" test and you will
immediately see the difference).
If font caching would be involved, the increase of the time with plot
complexity does not sound reasonable.

I am not sure how specific this is for 3.6.2 (compared to older versions
- colleagues say that they have less of an issue for 3.2.4 mingw, but I
do not have that available myself at the moment) or for msvc or for
windows in general...

Thanks for any hint,

Rolf