Don't fail matplotlib related doctests if DISPLAY is unset

[infinity0]

Debian (and I'm sure other distros)'s autobuilders run with unset DISPLAY, and the tests fail without this patch.

Component: interfaces

Author: Ximin Luo

Branch/Commit: u/infinity0/don_t_fail_plot_py_doctest_if_display_is_unset @ a017372

Reviewer: François Bissey

Issue created by migration from https://trac.sagemath.org/ticket/22441

[infinity0]

Branch: u/infinity0/don_t_fail_plot_py_doctest_if_display_is_unset

[infinity0]

Commit: a017372

[infinity0]

Author: Ximin Luo

[infinity0]

New commits:

a017372

Don't fail matplotlib related doctests if DISPLAY is unset

[videlec]

comment:4

I found weird that this modification has to be done only in two places... there are plots everywhere in the doctests! Why is that? Wouldn't it be better to set it automatically by the doctest engine?

[infinity0]

comment:5

Debian is on matplotlib v2 and Sage is still using v1.5 (see our status page).

I did initially also add it into other places, but only these two were needed in the end to fix the test failures. Perhaps it runs the other plots after running these tests, so the "good" settings have already been set by then?

Agreed that this is best done in the main code itself; probably not even in the doctest engine, but the Sage-matplotlib interface. I'm not certain where would be the best place though, I'm not familiar with the code.

[infinity0]

comment:6

It might be cleaner to patch matplotlib itself, I've filed a ticket on their side too.

[jdemeyer]

comment:7

I cannot reproduce this problem. I'm sure that many bots run Sage with DISPLAY unset, so there must be something more going on...

[infinity0]

comment:8

Looks like upstream rejected my suggestion. I can't reproduce this now on my main computer either, but I'll try it again in Debian's build-in-a-chroot program later.

env -i PATH=/usr/bin:/bin HOME=$HOME sage -c 'plot(x^2, (x,0,5))'

They did point me to the MPLBACKEND envvar so perhaps Sage's doctesting framework can set this envvar, assuming the "real cause" isn't something else.

[jdemeyer]

comment:9

Any further news? Can this be reproduced in vanilla Sage?

[fchapoton]

comment:10

see #23696 for matplotlib update

[saraedum]

comment:11

It appears that #23696 fixed this:

+# Make sure the agg backend is selected during doctesting
+# This needs to be done before any other matplotlib calls.
+import matplotlib
+matplotlib.use('agg')

So, can this be closed now?

[kiwifb]

comment:12

Yes, I didn't know there was a matching ticket to the debian patch. But this was definitely part of my target in #23696.

[kiwifb]

Reviewer: François Bissey

[videlec]

comment:13

closing positively reviewed duplicates