Currently, the ps backend positions eps images into an imaginary page (the smallest standard page size which can contain the figure size, which happens to be lettersized for the default figure size), setting the %%BoundingBox based on that positioning. But the eps file format doesn't actually record the full page size (that's not the intent of the eps format anyways), just the position of the drawing relative to an origin.

GhostScript is then used to rasterize the eps images during testing; it implicitly assumes a default page size of "letter" (https://ghostscript.readthedocs.io/en/latest/Use.html#choosing-paper-size) which just happens to match the page size selected above. So things are OK... except if the test figure is nondefault and actually bigger than lettersized; in that case ghostscript will just crop out whatever is out of the lettersized paper. Note that such an image comparison test won't fail; it will just fail to compare anything that's outside of the lettersize paper.

Instead, pass -dEPSCrop to GhostScript
(https://ghostscript.readthedocs.io/en/latest/Use.html#depscrop) which readjusts the papersize to match the eps bounding box.

Test e.g. with

import subprocess
from matplotlib.figure import Figure
for fs in [5, 10, 15, 20]:
    Figure(figsize=(fs, fs)).add_subplot().figure.savefig(f"test-{fs}.eps")
    subprocess.run([
        "gs", "-dNOSAFER", "-dNOPAUSE", "-dEPSCrop",
        "-o", f"test-{fs}.png", "-sDEVICE=png16m", f"test-{fs}.eps"])

(Noted while troubleshooting failed tests on mplcairo, which does not perform the centering -- but there were so few tests using eps that the difference in behavior was entirely hidden by the general mplcairo test tolerance, until some more tests were added in matplotlib 3.6.)

Not really a regression, but would still be nice to have this in 3.6.1 to make testing of mplcairo simpler (otherwise I'll just xfail some more tests in mplcairo...).

PR Summary

PR Checklist

Tests and Styling

• Has pytest style unit tests (and pytest passes).
• Is Flake 8 compliant (install flake8-docstrings and run flake8 --docstring-convention=all).

Documentation

• New features are documented, with examples if plot related.
• New features have an entry in doc/users/next_whats_new/ (follow instructions in README.rst there).
• API changes documented in doc/api/next_api_changes/ (follow instructions in README.rst there).
• Documentation is sphinx and numpydoc compliant (the docs should build without error).