Re: Improvements in subplot

[logari81]

On Thu, 2011-03-10 at 20:37 +0100, Konstantinos Poulios wrote:
> On Sun, Mar 6, 2011 at 8:16 PM, Ben Abbott <address@hidden> wrote:
> > On Mar 6, 2011, at 1:27 PM, logari81 wrote:
> >
> >> On Fri, 2011-03-04 at 21:10 +0000, bpabbott wrote:
> >>> On Mar 04, 2011, at 05:57 AM, Konstantinos Poulios
> >>> <address@hidden> wrote:
> >>>
> >>>> Hi all,
> >>>>
> >>>> I have implemented an idea that I had about subplot. You can try it
> >>>> by
> >>>> using the attached m-files.
> >>>>
> >>>> It is not finished yet, so expect some issues (e.g. when subplots do
> >>>> not have any titles or axes labels)
> >>>>
> >>>> However the basic concept, seems to work quite well.
> >>>>
> >>>> Some points that I could mention:
> >>>>
> >>>> 1. For both gnuplot and octave if the align_axes option is not
> >>>> present, "activepositionproperty" is set to "position". So we
> >>>> decline
> >>>> from ML, but I think this makes sense because MLs' subplots are not
> >>>> perfect either.
> >>>>
> >>>> 2. Only for fltk, I used listeners in order to notify all subplots
> >>>> when tightinset of one of them changes.
> >>>>
> >>>> 3. Only for fltk, I used the looseinset property in order to align
> >>>> axes in column rows and columns. The result is quite neat, see the
> >>>> attached png's. This is the reason for using
> >>>> "activepositionproperty"="position", otherwise looseinset would have
> >>>> no effect.
> >>>>
> >>>> 4. I have some concerns about the performance since there is a
> >>>> significant number of listeners involved. On my pc however it seems
> >>>> to
> >>>> perform quite well:
> >>>> ||octave:40> kostas_subplot_test
> >>>> ||1. Drawing a grid of 3x3 plot: 0.09 secs
> >>>> ||2. Add some data to the plots: 0.33 secs
> >>>> ||3. Add titles: 0.35 secs
> >>>> ||4. Add axes labels: 0.79 secs
> >>>>
> >>>> 5. For the moment this implementation works only for normalized
> >>>> units,
> >>>> so I have to add some units conversion inside a protected context.
> >>>> It
> >>>> is in my TODO list.
> >>>>
> >>>> I would like to have some comments and testing on this solution.
> >>>>
> >>>> @Ben, could you test the gnuplot part, which I am not very familiar
> >>>> with?
> >>>>
> >>>> Best regards
> >>>>
> >>>> Kostas
> >>>
> >>> I ran the subplot demos. Everything there looks ok to me. I'll run all
> >>> the demos over the week end and report back on the results.
> >>>
> >>>
> >>> Ben
> >>
> >> Hi Ben,
> >>
> >> actually I would like to send you an improved version but there is an
> >> issue that I can't solve:
> >>
> >> I add the listeners for each subplot as I described in my previous mail
> >> and everything works fine, until I plot something on the subplot axes.
> >> The plot command calls "set_defaults" which in turn removes all
> >> listeners.
> >>
> >> How could I make the listeners persistent? Do you have any clue?
> >>
> >> BR
> >>
> >> Kostas
> >
> > Unfortunately, much of the c++ side remains a mystery for me. Looking at 
> > graphics.cc:set_defaults, I don't see where the listeners are being deleted.
> >
> > I'd expect the listeners to remain in place, unless (1) they are removed by 
> > dellistener(), or (2) the axis object is deleted.
> >
> > If anyone else has some insight into what set_defaults is doing, the script 
> > below ....
> >
> > clf
> > axis;
> > fun = @() printf ("listener: active property = %s\n", get (gca, 
> > "activepositionproperty"));
> > addlistener (gca, "activepositionproperty", fun)
> > set (gca, "activepositionproperty", "position")
> > x = 0:0.1:10;
> > plot (x, sin(x));
> > get (gca, "activepositionproperty")
> > set (gca, "activepositionproperty", "outerposition")
> > get (gca, "activepositionproperty")
> >
> > ... should produce (I think?) ...
> >
> > listener: active property = position
> > ans = position
> > listener: active property = outerposition
> > ans = outerposition
> >
> > ... but because the listerner is deleted (accidentally?) when 
> > "plot(x,sin(x))" is run, I get ....
> >
> > listener: active property = position
> > ans = position
> > ans = outerposition
> >
> > There no need to apply Kosta's patch to test this behavior.
> >
> > Ben
> >
> 
> After having pushed the changes for supporting persistent listeners
> you can test the attached subplot version.
> 
> Regards
> 
> Kostas

Ben, could you test this changes now that we have circumvented the
issues with the crashes? If there are no regressions, performance issues
or other concerns, I would like to push this change.

Kostas