Repository navigation
Conversation
The upload action deprecated the singular `track` and warns on every run; it also rejects the build outright if both forms are set, so this is a swap rather than an addition. The workflow's own input stays singular: it passes exactly one track, and renaming it would churn release.yml and the dispatch UI for nothing. It now falls back to `internal` at the call site as well as in the input default, because the action uploads to *production* when it receives no track at all — not a default worth reaching by accident. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
[record-screenshots] The only link Coil answered started a favourite, which is the wrong tool for "put the app on screen" — and for a particular box, there was no way to ask at all. `coil://open` opens the app; `coil://open?box=<id>` opens it showing that box. A separate host rather than a play link with the target left off, because the two do different things and are worth telling apart in a note or an automation app: a play link reaches the box, this one never does. That also makes it the less privileged of the two — all it can ask for is what tapping the launcher icon already does, plus which of the user's own boxes comes up. `box` is optional here where a play link requires it: a folder path means nothing without the box it came from, but "just open the app" is a complete request. The id is checked against the configured boxes before use, since setActive writes straight to settings and a stale link would otherwise point the app at a box that does not exist. Settings hands the link out per box, for the same reason a favourite's menu does: the id is a UUID shown nowhere else, so `?box=` would be unusable otherwise. The share-sheet helper moved to ui/components to be shared with the favourites screen. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
[record-screenshots] Every string added since these files were first written had gone in as English only — the sleep timer, search, the details sheet, the library crawl, the links, and the settings rows added yesterday. 46 keys, missing in all four translated locales. Nothing reported it. Android falls back to the source locale silently, so each feature looked finished while three of the five launch locales showed English in the middle of a translated screen; `library/folders_de` had been showing "Search the library" over German tabs since the day it was recorded. MissingTranslation was configured as a warning and scrolled past. Also adds the CLDR "many" category to the French and Spanish plurals, which lint had been asking for on the three freshness labels as well. Drafts, marked as such, in the same standing as the rest of these files: they still need a fluent speaker. Absence was the worse failure, because a draft gets reviewed and an absence does not — AGENTS.md now says a new string lands in all five locales in the same change, and names the lint check that proves it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Picking a box, adding a box and editing the current box sat next to each other in settings as rows of the same shape, so the section read as three overlapping ways to say "box". Box management is now its own screen with a page per box, behind one "Manage boxes" row. Both screens address a box by id rather than through "the active box", so a box that is not active can be renamed or re-addressed without switching to it first — previously impossible, since every field described whichever box was active. Switching stays in the top bar only; the library actions stay in settings, because they run against the active box. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A Phoniebox serves artwork over plain HTTP from its own web server, and Android refuses cleartext by default at targetSdk 28 and up, so every cover request was rejected before it left the phone. Verified against a real box that the protocol side was never at fault: playerstatus.file -> get_single_coverart -> the HTTP fetch all answer correctly. It has to be a base-config rather than a domain-config, because the box address is entered at runtime and this file is fixed at build time. Two things hid this through an earlier hardware run: the ZMQ sockets are raw TCP and so were never subject to the policy, leaving everything but covers working; and the cover error branch draws the same placeholder as a song with genuinely no artwork. The image loader now carries a DebugLogger in debug builds so the next such failure names its reason. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The site was a single Markdown page from before there were screenshots or a settled feature set. It now has a hero, the twelve features, phone and tablet galleries, a lightbox and a links panel. Built on emilbaehr/automatic-app-landing-page (MIT). Vendored rather than used via remote_theme, because the theme keeps index.html and main.scss at its repo root while remote_theme only pulls _layouts, _includes, _sass and assets; its licence notice travels along in docs/pages/_THEME-LICENSE.txt. The theme is written for iOS apps, so the parts that assumed one had to go. The iPhone body and its notched clip-path were sized 1125x2436 and squashed Coil's 1233x2460 shots; it is now a CSS Android bezel that takes its height from the image. The unconditional App Store badge and apple-itunes-app banner became a Play badge and a Releases button. The itunes.apple.com lookup that filled in icon, name and price at page load is gone, along with the jQuery it needed. The tablet strip is a curated three. The 10-inch player and sleep timer use the width properly, but library, favourites and search still stretch the phone layout across 1280 dp, and showing those advertises the weakest part of the tablet build. flattenStoreAssets now copies all three sets to the site so that choice stays a _config.yml edit. ci.yml gains a Pages build, since pages.yml only runs on main and a broken template would otherwise reach the live site unnoticed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
"Show controls" offered two settings and both put a notification up sooner or later. A third, "Never", creates no media session at all. The mode is enforced in MediaSessionBinder rather than the service: a bound service outlives its own stopSelf, so dropping the binding is what actually keeps media3 from posting a notification on the next state == play. The service stops itself on the same setting, which is what takes an existing notification down mid-playback. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The check could never pass. `github-pages build` sets `exclude: [CNAME]` itself whenever the site's own `exclude` is Jekyll's default, so the official builder strips docs/pages/CNAME from the output on purpose. Removing the upstream theme's `exclude` block is what put the config back on that default and armed the rule. The custom domain is unaffected, and never depended on that file: the Pages API reports cname=coilforphoniebox.app with build_type=workflow, so the domain comes from the repository settings and nothing reads a CNAME out of an Actions artifact. A local `jekyll build` has no such rule and does copy CNAME, which is why this only surfaced in CI. In its place, check that the galleries and the hero frame actually rendered and that no Liquid tag was left unprocessed — failures a missing-file test would not have caught. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The third session-mode option landed in bb71543 without its pictures, so verifyRoborazziDebug failed on settings_ and settings_lower_ for phone and tablet. The diff is the new row and the shift below it; nothing else on those screens moved. Small is untouched — at 360x640 the group sits below the fold in both goldens either way. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
An ordinary build never compares the goldens — `./gradlew test` only checks that every screen still composes — so a stale one survives locally and CI is the first place it shows up. That is a bad moment to find out during a release. The new step works out whether the release touched UI or user-visible strings without touching the screenshots folder, and treats that as reason enough to run verify before committing. 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.