Repository navigation
[Bug]: check_figures_equal decorator doesn't respect new figure properties set in test #24875
Description
Bug summary
We use the check_figures_equal in testing to compare figures. This decorator uses two figure instances fig_test and fig_ref for comparison. If we set any figure properties (like facecolor) in the test, they don't seem to be propagated when saving the actual figure.
Ran into this while writing a test for #24816 (cc @anntzer, for reference)
Code for reproduction
@check_figures_equal(extensions=["png"])
def test_test(fig_test, fig_ref):
ax_test = fig_test.add_subplot()
ax_test.set(xticks=[], yticks=[])
fig_test.set_facecolor("teal")
ax_ref = fig_ref.add_subplot()
ax_ref.set(xticks=[], yticks=[])
fig_ref.set_facecolor("none")Actual outcome
This test passes, which suggests that the two figures created are equal but they actually aren't since the facecolor for both figures are set to be different.
Expected outcome
This test shouldn't pass and the figures should have facecolors "teal" and "none" respectively.
Additional information
No response
Operating system
Arch
Matplotlib Version
3.6.0.dev1889+gc59a7fa8db.d20230103
Matplotlib Backend
No response
Python version
3.10
Jupyter version
No response
Installation
None
Activity
anntzer commented on Jan 3, 2023
I haven't actually checked, but are you sure this is not due to a combination of a) tests defaulting to running with the classic (pre-mpl2) style and b) classic style setting savefig.facecolor = savefig.edgecolor = w? (The new default style sets savefig.facecolor = savefig.edgecolor = auto, which means "don't change it".)
chahak13 commented on Jan 4, 2023
It is indeed because of the default style being classic. I added an option to change style to the decorator and it worked with "mpl20". Shouldn't check_figures_equal also have an option like image_comparison to set style?
chahak13 commented on Jan 4, 2023
Also, how is the default determined? I'm not changing anything for it to run in classic or otherwise.
anntzer commented on Jan 4, 2023
Just decorate with @mpl.style.use(...) outside of the test decorator?
The default is determined by matplotlib.testing.conftest.mpl_test_settings.