Backport: SVG output: incremental ID scheme for non-rectangular clip paths.

This backports an upstream patch to fix an output nondeterminism (e.g. build reproducibility) bug - it is available in v3.10 of matplotlib upstream, but is relatively straightforward to backport to v3.8.3 also.

Closes #1065124.

Aside / Build-Dep modification explainer

The rendering of a TextPath entity introduced by the determinism patch causes a subtle build-time bug during autopkgtests: dvips produces intermediate DVI files that include Encoding vector references for the tcss3583 font -- because that font appears in dvips-all.enc as distributed in texlive-base. the fact that matplotlib produces and processes TeX that includes \usepackage{type1ec} -- without specifying a 10-point font size! -- means that large-sized text may attempt to use fonts such as tcss3583 that are not available in cm-super-minimal

(the minimal variant of the package only contains 10* font files).

However: we do not currently install the referenced font file itself during package build, and matplotlib fails to render the PostScript version of the deterministic test diagram as a result of that.

In this merge request, commit 80b52c68 adds a workaround: it migrates the build-dep on cm-super-minimal to cm-super so that the TCSS3583 font is available when the test suite runs. However, I'll try to learn more about the problem and report it as a bug upstream in matplotlib; there may be a way to handle this condition as a non-fatal error.

Edit: markup fixup for malformatted subheading in description

Edit 2: provide a more accurate problem description

Edit 3: improve the formatting of these edit footnotes

Edit 4: update a reference in the closing paragraph to commit 1375a4fa -- it has been superseded by 80b52c68 (with an explanatory commit message)

changed the description

The rendering of a TextPath entity introduced by the determinism patch causes a subtle build-time bug during autopkgtests: dvips produces intermediate DVI files that include Encoding vector references for the tcss3583 font -- because that font appears in dvips-all.enc as distributed in texlive-base.

This appears to be inaccurate: in fact, the reference to tcss3583 is introduced by a make_tex method that writes TeX output including \usepackage{type1ec}.

There is a comment to mention that doing so pulls in cm-super: https://salsa.debian.org/python-team/packages/matplotlib/-/blob/2558456d2550629c34553ce256c6158d377b1b1a/lib/matplotlib/texmanager.py#L84-86

...however: in the cm-super package README.Debian file, there is a note (well, a big warning actually) that font-size must be fixed to 10pt when using cm-super-minimal:

https://sources.debian.org/src/cm-super/0.3.4-17/debian/README.Debian/#L13-L18

The intermediate .tex files do not do this; they currently emit content that looks like this:

documentclass{article}
...
\usepackage{type1cm}
\usepackage{type1ec}
\usepackage{type1ec}
\usepackage{type1ec}
...
{\sffamily °} 
\end{document}

changed the description

added 1 commit

• 80b52c68 - Build-Deps: migrate to `cm-super` to provide `tcss3583` font

Compare with previous version

@detiste-guest @stuart could we include this for trixie? It's OK if not!

We could even try to have 3.10 in trixie ...

changed the description

We could even try to have 3.10 in trixie ...

...and that's appearing more likely, with the upload of 3.10 to experimental :) (thanks, @nilesh!)

I think that that could (and should) obsolete this 3.8 backporting approach, so I'll go ahead and close this.

(if for some unexpected reason we do want to backport this to earlier releases, that's fine by me - but I think it is probably simplest to await 3.10)

closed