Re: [Gnuplot-py-users] hardcopy termdef defnition for pdf

Michael Haggerty wrote:

> Jim Kleckner wrote:
> 
>> I added the code below to termdefs.py to allow hardcopy
>> to work with pdf.
> 
> Thanks, I just added this to CVS.  Sorry I didn't get it into the release.


Great, thanks!


>> I also found that the temporary file was being removed before
>> the gnuplot process was done with it.  Adding a sleep() of a
>> few seconds made it work but that was kind of a kludge.
>> Adding a g.gnuplot.gnuplot.close() worked fine since it
>> causes a wait for the popen() process to exit (if not using
>> inline).  I wonder if there is some way to put this into
>> the __del__() function at a place where this would happen
>> naturally?  I tried the obvious on Gnuplot but it didn't work.
> 
> The temporary-file removal problem is discussed in FAQ.txt.  Under what 
> circumstances exactly did calling close() help?  It would probably make 
> more sense to add a close() method to the GnuplotProcess objects because 
> its implementation is likely to be platform-specific.  Then I guess 
> Gnuplot should call the close() method in its own __del__ method.  Hmmm, 
> I guess this helps by ensuring that the pipe is closed before the 
> plotitems are destroyed.
> 
> OK, I just implemented this in CVS, too.
> 
> Thanks for your note and suggestions!
> 
> Michael
> 


Yes, when I upgraded to 1.7 CVS I noticed that cygwin
implemented inline by default now and that is the strongest
way to guarantee that the data will be synchronized.  I ended
up switching everything over to that.

I switched back over to "not inline" and it still fails when

you reuse an existing plot process.  The race seems to be that
when you perform a new plot (without closing the process)
and the piped command hasn't finished processing the last file,
it can get deleted by the next rapid fire g.plot() request that
replaces the plot items.  A total hack could be to keep an
invisible list of plot items not in the queue that gets set to
None at close time resulting in file deletion.  Quite a hack.
Proabably better to tell people to just use inline or create
a new plot object for each plot.


Thanks!
Jim