Upgrade matplotlib to 1.2.1

[jason]

Matplotlib 1.2.1 is released now: ​http://matplotlib.1069221.n5.nabble.com/ANN-matplotlib-1-2-1-release-td40752.html

spkg: ​http://sage.math.washington.edu/home/jpflori/matplotlib-1.2.1.spkg

apply: trac_13693-part1.patch​

[jhpalmieri]

I'm trying this out. A few quick comments:

• In SPKG.txt, the line === matplotlib-1.1.0 (Jason Grout, 09 Nov 2012) === should say 1.2.0.
• The file patches/mkdir-race-condition.patch should be removed from the repository (according to hg status, the repo is not clean right now).
• I'm getting doctest failures in the plot directory:

  The following tests failed:
  
  	sage -t  devel/sage-main/sage/plot/colors.py # 8 doctests failed
  	sage -t  devel/sage-main/sage/plot/graphics.py # 1 doctests failed
  	sage -t  devel/sage-main/sage/plot/plot.py # 2 doctests failed
  	sage -t  devel/sage-main/sage/plot/plot3d/plot_field3d.py # 1 doctests failed
  

  The ones in colors.py are all of the form

      sage: get_cmap('jet')
  Expected:
      <matplotlib.colors.LinearSegmentedColormap instance at 0x...>
  Got:
      <matplotlib.colors.LinearSegmentedColormap object at 0x11103e8d0>
  

  and the ones in graphics.py are similar. These are easy to fix, of course. Other failures:

  File "/Users/palmieri/Desktop/Sage_stuff/sage_builds/sage-5.4.rc4-IPYTHON/devel/sage-main/sage/plot/plot.py", line 1176:
      sage: len((q1).matplotlib().axes[0].legend().texts) # used to raise AttributeError
  Expected:
      1
  Got:
      2
  **********************************************************************
  File "/Users/palmieri/Desktop/Sage_stuff/sage_builds/sage-5.4.rc4-IPYTHON/devel/sage-main/sage/plot/plot.py", line 1186:
      sage: len(p1.matplotlib().axes[0].legend().texts)
  Expected:
      1
  Got:
      3
  

  Why is this failing now? Was there a change in the format of the legend? The error in plot_field3d.py is ValueError: Colormap red is not recognized. I don't know what to do about this one. I haven't run full doctests, just on the plot directory. I'll let you know if I find anything else.

[jason]

Wow, sorry, that was really sloppy on my part. This is definitely needs work right now.

[jason]

Those printing issues are weird. I wonder if that's from the upgrade to IPython 0.13.1?

[jason]

I've uploaded a new spkg that should fix the two issues with the spkg.

[jhpalmieri]

I think I get the printing issues with both the old and the new IPython. See /scratch/palmieri/sage-5.4.rc4-13693/ on sage.math for a build with a stock 5.4.rc4 plus this spkg.

[ddrake]

I hope we can get this into Sage -- version 1.2 has ​support for PGF/TikZ output which opens up some pretty amazing opportunities with SageTeX. (Not that I will have time to implement that in the near future, but getting this upgrade in is a necessary first step.)

[kcrisman]

See also ​streamplots - for #10775.

[fbissey]

Definitely would be nice. I had seen the failures before in sage-on-gentoo before forcing 1.1.0. I will see if I can dedicate some time to it this week.

[jhpalmieri]

Regarding the failures: I can easily fix all of them except the ones in plot.py:

        sage: q1 = plot([sin(x), tan(x)], legend_label='trig')
        sage: len(q1.matplotlib().axes[0].legend().texts) # used to raise AttributeError
        1

    ::

    Make sure that we don't get multiple legend labels for plot segments
    (:trac:`11998`)::

        sage: p1 = plot(1/(x^2-1),(x,-2,2),legend_label="foo",detect_poles=True)
        sage: len(p1.matplotlib().axes[0].legend().texts)
        1

For the first one (q1), with the new spkg there are two legend labels ("trig" and "None"), where there used to be just one. For p1, there are now three labels, in direct contradiction of the text. It looks like the picture at #11998, except the labels are "None", "None", and "foo". I don't know how to fix these. See my patch for the other fixes.

[jpflori]

I was just doing the same thing this evening to try random Cygwin stuff.

I guess we should look for 1.2.1.

[jpflori]

Updated spkg at: ​http://boxen.math.washington.edu/home/jpflori/matplotlib-1.2.1.spkg

[jpflori]

Not sure if it's the update wihch changed something but I have no failures in plot.py on my 5.9.beta1 install.

I'm putting this back to needs_review though I could not run "make ptestlong" yet cause the install I have access to is kind of borken because of ATLAS black magic.

[jhpalmieri]

Replying to jpflori:

Not sure if it's the update wihch changed something but I have no failures in plot.py on my 5.9.beta1 install.

Same for me, too. I'll try make ptestlong on a few machines to see what happens.

[jason]

Has anyone manually checked the images connected to the failing doctests from before, just to make sure we're not missing a problem?

[jpflori]

I had a look at the q1 one to spot the problem but it looked fine (blue sine and other curve, axes, little box with a blue segment and trig, no None anywhere) and was confused and then indeed there was no failing doctest.

[jhpalmieri]

I also looked at the examples by hand, and they looked fine (as opposed to with the 1.2.0 spkg). make ptestlong passed on two platforms: boxen.math.washington.edu and an OS X 10.8.3 machine.

So I'm happy with the spkg. If people are happy with my patch, can we give this a positive review?

[fbissey]

I think the patch is all fine. I had to check if it was the intended behavior for plot_field3d.py (monochrome 3d field, I guess it can be useful sometimes).

[jdemeyer]

I'm assuming the patch should be applied...

[fbissey]

Yes I should have done the clean up of the description.

[jdemeyer]

This is very wrong and should be fixed:

        assert(cm is not None) 
    except (TypeError, AssertionError, ValueError): 

AssertionError should never be handled in an exception block, unless you have very good reasons to do so. The obvious solution is replacing assert() by something else.

I know this problem isn't caused by this patch, but since it is easy to fix, you should fix it.

[jdemeyer]

For example, you could do:

try:
    cm = get_cmap(colors)
except (TypeError, ValueError):
    cm = None
if cm is None:
    ...

[jhpalmieri]

Okay, here's a new version of the patch. The only change is this:

sage/plot/plot3d/plot_field3d.py

@@ -72 +72 @@
     try:
         from matplotlib.cm import get_cmap
         cm = get_cmap(colors)
-        assert(cm is not None)
-    except (TypeError, AssertionError, ValueError):
+    except (TypeError, ValueError):
+        cm = None
+    if cm is None:
         if isinstance(colors, (list, tuple)):
             from matplotlib.colors import LinearSegmentedColormap
             cm = LinearSegmentedColormap.from_list('mymap',colors)

[jhpalmieri]

fixes for failures except for plot.py

[jhpalmieri]

I'm going to restore the positive review, also; you can view my change as a positive review of Jeroen's suggestion...

[nthiery]

As a followup: the new version of matplotlib seems to be causing an issue with saving graphs in pdf. See:

​https://groups.google.com/forum/?fromgroups=#!topic/sage-devel/6l9Z4y5rCuM