What
check_plugin.py fails for ledmatrix-weather on a clean main, with no local changes:
[FAIL] 64x32 almanac golden drift: 267px (max Δ=255)
[FAIL] 128x32 almanac golden drift: 810px (max Δ=255)
[FAIL] 128x64 almanac golden drift: 747px (max Δ=255)
[FAIL] 256x32 almanac golden drift: 810px (max Δ=255)
The other 36 checks pass, and only the almanac mode drifts. I verified this is pre-existing by stashing my branch and re-running against main — the failures and pixel counts are identical, so it is not caused by #392.
max Δ=255 means the differing pixels are fully on in one image and fully off in the other, which points at glyph or icon coverage rather than a colour shift.
Probably the same cause as three others
Three plugins already have open issues for golden drift on main with no local change:
That is now four. A per-plugin cause four times over is less likely than one shared one — the obvious candidate is the Pillow / FreeType version the goldens were generated against versus the one running now. Related: #383 found that the documentation renderer picks a different caption font per host, which is a different mechanism but the same underlying fragility about what "the same render" means across machines.
Suggested next step
Rather than regenerating four sets of goldens and hoping, it would be worth establishing what the goldens were generated on. If they came from a different Pillow major version, pinning Pillow for the harness (or recording the version alongside the goldens) fixes all four at once and stops it recurring. If the versions match, then each is a real regression and deserves separate investigation.
I have deliberately not regenerated any goldens.
What
check_plugin.pyfails forledmatrix-weatheron a cleanmain, with no local changes:The other 36 checks pass, and only the
almanacmode drifts. I verified this is pre-existing by stashing my branch and re-running againstmain— the failures and pixel counts are identical, so it is not caused by #392.max Δ=255means the differing pixels are fully on in one image and fully off in the other, which points at glyph or icon coverage rather than a colour shift.Probably the same cause as three others
Three plugins already have open issues for golden drift on
mainwith no local change:of-the-daygeochronchristmas-countdownThat is now four. A per-plugin cause four times over is less likely than one shared one — the obvious candidate is the Pillow / FreeType version the goldens were generated against versus the one running now. Related: #383 found that the documentation renderer picks a different caption font per host, which is a different mechanism but the same underlying fragility about what "the same render" means across machines.
Suggested next step
Rather than regenerating four sets of goldens and hoping, it would be worth establishing what the goldens were generated on. If they came from a different Pillow major version, pinning Pillow for the harness (or recording the version alongside the goldens) fixes all four at once and stops it recurring. If the versions match, then each is a real regression and deserves separate investigation.
I have deliberately not regenerated any goldens.