[Bug]: EPS export shows hidden tick labels when using tex for text rendering

[yschroeder]

Bug summary

When setting text.usetex: True, the EPS export produces spurious axis labels which are not visible in the PDF export.

This seems somewhat similar to #10163.

Code for reproduction

import matplotlib.pyplot as plt

# using tex shows spurious labels on the x axis
plt.rcParams.update({"text.usetex": True})

fig, axs = plt.subplots(1, 3, sharey=True)

# add a big axes, hide frame
hidden_ax = fig.add_subplot(111, frameon=False)

# hide tick and tick label of the big axes
plt.tick_params(labelcolor='none', top=False, bottom=False, left=False, right=False)

# comment in this workaround
# as suggested in https://github.com/matplotlib/matplotlib/issues/10163/
# hidden_ax.xaxis.set_major_locator(plt.NullLocator())
# hidden_ax.yaxis.set_major_locator(plt.NullLocator())

# save figures
fig.savefig("test.eps")
fig.savefig("test.pdf")

Actual outcome

Expected outcome

Operating system

Arch

Matplotlib Version

3.4.3

Matplotlib Backend

Qt5Agg

Python version

3.9.7

Jupyter version

No response

Other libraries

texlive-core 2021.58710-2
texlive-bin 2021.58686-3

Installation

Linux package manager (Debian/Fedora/etc.)

Conda channel

No response

[anntzer]

Looks like the same fix as in #10166 applies, i.e.

diff --git i/lib/matplotlib/backends/backend_ps.py w/lib/matplotlib/backends/backend_ps.py
index c44f89c638..35c61b08f2 100644
--- i/lib/matplotlib/backends/backend_ps.py
+++ w/lib/matplotlib/backends/backend_ps.py
@@ -593,6 +593,9 @@ translate
     @_log_if_debug_on
     def draw_tex(self, gc, x, y, s, prop, angle, *, mtext=None):
         # docstring inherited
+        if self._is_transparent(gc.get_rgb()):
+            return  # Special handling for fully transparent.
+
         if not hasattr(self, "psfrag"):
             self._logwarn_once(
                 "The PS backend determines usetex status solely based on "

seems to fix the problem?

Probably the test_transparency intrroduced in #10166 could be parametrized to also test the usetex=True case.

[yschroeder]

I looked at #10166 yesterday and also had the hunch that your proposed check is missing in the draw_tex() function. However I was still in the process of setting up a matplotlib development environment, you were faster with the patch 😄

I have the dev environment set up now and can confirm that test_transparency() exposes the bug by adding plt.rcParams.update({"text.usetex": True}) as the first line in that test. By adding your patch to the PS backend, the modified test that exposed the bug is green again.

I can further confirm that my original problem is solved as well by the patch.

So overall, your patch fixes the problem and a modified version of test_transparency() is able to avoid a regression.

Will you prepare a PR for this?

[QuLogic]

Usually when @anntzer adds that label, it's because he hasn't the time to do the full PR with tests/docs/etc. So feel free to open one yourself.

[yschroeder]

OK, I went ahead and created a PR.

[jklymak]

Closed by #21470