COMMIT_LIST only rendered the bare subject line per commit, with no
way to jump to that specific commit — only the triggering commit
(Source:) had a link. Each line now reads
"- [<short-sha>](<repo>/commit/<sha>) <subject>", matching the same
link style already used for Source.
adguard-cidre switched its release changelog from a changed-files list
to a commits list (with --- dividers) after the previous commit here
was written from an already-stale read of that file. Also drops the
glob-based smoke test (never actually part of the real template, only
of nvidia-stock-bot's own drift) for an explicit per-file check
matching the canonical error-handling shape, and aligns the remaining
step names/commit message wording.
Pin Trivy to 0.74.0 instead of latest, make the deprecation check
non-blocking (warning only), log the release API's HTTP status code so
a failure is actually visible in the CI logs, drop the redundant
Docker socket mount on the Trivy steps, and switch the version-bump
push to the same URL-embedded-token approach and extraheader key the
other repos use. Also generalizes the smoke test to every top-level
.py file instead of a hardcoded list, and enriches the release body
(changelog header, source links, changed files) to match the other
repos' output.
Adds the weekly Monday 6am schedule trigger too, so a Trivy scan still
runs even on a week Renovate finds nothing to bump.
83 tests (94% coverage of src/py) covering the builder pipeline
(gallery sync, HTML/CSS generation, image processing, full site
build) and the Flask webui (routes, uploads, theme/font management).
Dockerfile gains a `test` stage (pytest) between the wheel builder
and the prod image, and pins the alpine base to a full patch tag so
Renovate can classify updates. CI workflow builds, smoke-tests, runs
the suite, scans with Trivy, and publishes/releases on merge to main,
following the same pipeline already running on adguard-cidre.
Image.open() reads the source's ICC profile into img.info, but that dict resets to empty after convert() or resize(), and save() was never passing icc_profile anyway. DCI-P3 and Rec.2020 photos were silently falling back to sRGB interpretation once processed.
Djeex
Djeex