recent snapshot and fltk plotting

[Michael D Godfrey]

I ran some plotting tests (graphics_toolkit fltk) on the snapshot just 
before
3.5.91.  This was a severe stress test. The main script generates
about 100 plots, using system() and eval() calls and running
through the NX remote desktop (which slows down plotting quite a lot).

The results did not show any significant regressions.  But, the
longstanding timing problem is still there:  without any pause()
statements random print() commands do not generate the intended xxx.pdf
files. Typically about 3 to 5 out of about 100 plots were missing.  
Rerunning
the same script produced a different set of missing plots. After some
experimenting, a few pause(1) statements at the right place in the output
loops caused all plots to be reliably and correctly produced. Also,
the print commands actually were executed through eval(xxx), so that
added to the test.

One suspicious behavior was that the plots that appeared, and were then
replaced by the next plot, were often incomplete or garbled before they
disappeared.  However, the output PDF was always correct.
In any case, there is still a timing window in the fltk plotting
which has the effect of causing the intended .pdf to not get written.  This
seems only to  appear under conditions (NFS mounted filesystems, DSL, 
NX,...)
which slow down graphics data transfers.  In the past I always had to
break this script into short pieces in order to get all the plots, or 
work on
a machine which ran all the code directly through X11 or at least on a high-
speed connection.

I have been unable to find a way of reliably reproducing this problem and
therefore cannot put together a useful bug report.

So, I would say that the current release, in this respect, is as good 
(better actually)
as any previous version.  This code had not been run for about 18 months and
it all worked exactly without any necessary change except for the 
pause() statements.

Michael