Conversation
A Phoniebox library is largely ripped CDs and home-made folders, so much of it carries no embedded artwork — and the favourites tab and album grid were walls of identical grey discs, no use to the child this app is pointed at, who navigates by picture rather than by reading. Thirty-two covers ship in res/drawable-nodpi and one is assigned per folder, album or track name, so the same name draws the same picture on every screen and after a reinstall. Artwork the box does have still wins. nodpi rather than plain drawable because these are art scaled to whatever slot they land in: in drawable/ they would be treated as mdpi and upscaled some 3.5x in memory on an xxhdpi phone. The hash is FNV-1a, not String.hashCode. The latter is stable across hosts and versions, which is the property that matters, but it is not mixed: modulo 32 keeps only its low five bits, and for the runs a library is full of (Folge 01, Folge 02, ...) those move with the last character. Forty such folders come out as 14 distinct covers in a visibly repeating pattern against FNV-1a's 28. The mapping is frozen and pinned to values by FallbackCoverArtTest: changing the hash, the folding or the size of the set reshuffles artwork users have come to recognise, and no golden would notice — the fixtures would simply re-record. A null cover URL had meant two things: "no artwork" and "nobody has finished asking yet". Drawing a stand-in on every null would have put a picture on screen a second before the real cover replaced it, on every track change and across a whole album grid at once. CoverArt now takes coverPending, supplied by PlayerRepository for the player and mini player and by a settled-id set in the favourites and library view models. A stream has no folder whatever PlayerStatus.folder says — that property is the file path minus its last segment, which for a stream URL is the server, so every station on one host would have shared a picture. coverNameOf drops to the album and artist tags there. FakeCoverArt now serves these same covers instead of drawing its own, so the goldens and the store screenshots show what the product shows. That gives up the 3:2 landscape source that let a golden fail visibly on a stretched or letterboxed cover; those regressions still fail the test, since a golden is a byte comparison, but reading the diff is now the way to see what went wrong. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The favourites shot no longer has an empty tile in it: the folder the box has no artwork for now draws a stand-in like the rest. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The goldens are baselines a maintainer accepts rather than a contract the code must satisfy, so "the UI changed" is information rather than a defect. A golden can also differ for reasons unrelated to the change under review — a Robolectric or Skia bump, an image decoder that rounds differently on another host — and a check that goes red for those becomes a check that gets clicked past. continue-on-error leaves the job green; a PR annotation and a step summary say what moved and how to accept or reject it. The diff upload now keys on the step's outcome rather than failure(), because with continue-on-error the job never enters a failed state and the triptych would otherwise be silently dropped — the one artifact the job exists to produce. This changes the reporting only. Screenshots was never among main's required status checks (Build, Unit Tests, Lint, Analyze), so a mismatch could not block a merge before this either. The cost is not hedged in the comments or in AGENTS.md: nothing else in CI looks at pixels, so a genuine visual regression now merges green unless somebody reads the warning. If that stops happening, the honest fix is to make the job blocking again rather than keep a check nobody acts on. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.