Plugin
- Plugin id: christmas-countdown
- Plugin version: 1.0.2
Pre-existing on main — confirmed by stashing unrelated changes and re-running, giving identical figures.
Describe the bug
[FAIL] 64x32 golden drift: 177px (max Δ=255)
[FAIL] 128x32 golden drift: 244px (max Δ=255)
[FAIL] 64x64 golden drift: 177px (max Δ=255)
[FAIL] 96x48 golden drift: 244px (max Δ=255)
[FAIL] 128x64 golden drift: 244px (max Δ=255)
[FAIL] 256x32 golden drift: 244px (max Δ=255)
[FAIL] 128x96 golden drift: 244px (max Δ=255)
[PASS] 256x128 (golden ✓)
Two clusters — 177px on the two 64-wide panels, 244px on everything else — with 256×128 the lone pass. max Δ=255 means at least one pixel flipped between full-off and full-on rather than shifting slightly.
What I can and cannot say
The 64-wide split lines up with the XMAS / CHRISTMAS abbreviation boundary, so text rather than the tree looks like the likely difference. But that is inference from the failure pattern, not something I have confirmed.
I have not regenerated the goldens, for the same reason as #371 and #375: from here I cannot distinguish "the goldens are stale" from "the goldens are correctly catching a regression", and this plugin scales a PNG with Image.Resampling.LANCZOS, which is exactly the kind of operation whose output can differ between Pillow versions. Regenerating from my environment (Pillow 11.3.0) risks baking that in.
That is now three plugins with golden drift on main — this, #371 (of-the-day) and #375 (geochron). If they all pass in your environment, the common cause is likely a dependency-version difference rather than three separate regressions, and it would be worth pinning the version the goldens were generated against somewhere the harness can check.
Found while writing the README in #378.
Plugin
Pre-existing on
main— confirmed by stashing unrelated changes and re-running, giving identical figures.Describe the bug
Two clusters — 177px on the two 64-wide panels, 244px on everything else — with 256×128 the lone pass.
max Δ=255means at least one pixel flipped between full-off and full-on rather than shifting slightly.What I can and cannot say
The 64-wide split lines up with the
XMAS/CHRISTMASabbreviation boundary, so text rather than the tree looks like the likely difference. But that is inference from the failure pattern, not something I have confirmed.I have not regenerated the goldens, for the same reason as #371 and #375: from here I cannot distinguish "the goldens are stale" from "the goldens are correctly catching a regression", and this plugin scales a PNG with
Image.Resampling.LANCZOS, which is exactly the kind of operation whose output can differ between Pillow versions. Regenerating from my environment (Pillow 11.3.0) risks baking that in.That is now three plugins with golden drift on
main— this, #371 (of-the-day) and #375 (geochron). If they all pass in your environment, the common cause is likely a dependency-version difference rather than three separate regressions, and it would be worth pinning the version the goldens were generated against somewhere the harness can check.Found while writing the README in #378.