changelog/v<major>/<major>.<minor>.x/<version>.md for each of the 42 releases from v1.0.0 to v4.7.1, and changelog/beta/ for pre-releases (none has been published; its README says how one would be filed). CHANGELOG.md becomes the index: what the layout is, [Unreleased] exactly as it was, and a table per major version (version, date, edition) that links each file. Each file is its section, copied byte for byte except that its heading is the file's H1 (so the sub-headings move up one level, outside code fences), relative links gain ../../../ so they name the same repository paths, and the --- separators between sections are gone. Checked by reversing all three changes on every file and comparing with the original section: 40 of 40 equal. Two releases had no section: v1.0.0 (its GitHub release notes, which the changelog pointed to as "Initial stable") and v1.1.3 (its tag message). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
6.1 KiB
6.1 KiB
v4.4.0 – 2026-10-23 (The "Headroom" Edition)
The data pipeline grows with the catalogue again. With almost 10,000 games
and a few hundred more every day, the jobs that asked Steam about one game at
a time had run out of time; they now read Steam's store API fifty games a
request, wait their turn instead of cancelling each other, and stop in time to
save. The release also carries the tooling merged since v4.3.3 - TypeScript
7, Node 26 and Python 3.15, SonarCloud in CI, the desktop updater fix - and a
certifi bump. The desktop and Android apps stay at 2.2.1, with no new
build. Repo public-facing version 4.3.3 → 4.4.0.
🧰 Pipeline
- The data jobs scale with the catalogue again. With almost 10,000 games
and a few hundred more every day, three jobs that asked Steam about one
game at a time had reached GitHub's six-hour limit: on 2026-10-23 Update
Reviews was stopped at exactly six hours with nothing saved. The full
assessment, with every measurement, is
docs/plan/18-pipeline-scale.md. - Reviews come from Steam's store API, for every game, every day. Fifty games a request, about two minutes for the catalogue, instead of every two days and five and a half hours. The percentage is now Steam's own, which rounds down: about 4,000 games show one point less once (Counter-Strike 2 "85%" instead of "86%"), exactly as on their store pages. A summary that says 0 reviews is double-checked game by game before it replaces a percentage.
- New games get their reviews again. Steam's per-game reviews endpoint has answered with a redirect to the store front since 2026-10-23, so the 38 games added that day were listed with "N/A". A new game now takes its summary from the store API when that endpoint gives none, and a game never read becomes "No reviews" until Steam counts its first one.
- The daily purge needs two answers to remove a game. Steam's store API screens every game in a few minutes, and a game is removed only when appdetails agrees; "unavailable" is asked twice, five minutes apart. A run that would remove more than 100 games, or flags more than 500, stops before writing anything. It also lists earlier removals that Steam has since undone. It runs daily now, instead of a five-hour sweep on Mondays.
- Check Dead Links is retired. In its whole life it never found a dead
link: Steam sends a delisted game's page to the store front, which answers
200. The purge's screen sees delisted games (
unlisted) instead. - No run is cancelled for waiting. The data jobs share one queue that
used to hold a single waiting run: a third job cancelled it, and Update
Reviews lost five of ten runs that way. The queue now holds them all, first
in, first out, and every job checks out the latest
mainwhen it starts - Auto Update JSON failed four days in ten from working on the data as it was when the run was queued. - Every long job stops in time to save. Each workflow gives its job a time budget below its timeout: past it the job saves and commits what it did instead of being stopped with nothing, and it warns before the budget gets tight. The leaderboards skip the pauses meant for store pages (Top Online ~27 minutes instead of 48). The store-data and all-time-peak rotations stretch their cycle as the catalogue grows instead of their daily runs.
- Force Re-fetch runs in parts (
1/8…8/8, queued one after another), and a game whose fetch fails keeps its old data.
🧰 Tooling
- Desktop releases keep reaching the updater. tauri-action v1 writes each
file's API address into
latest.json, which the app's update check refuses, so desktop 2.2.1 had to be fixed by hand before installed apps were offered it. A new step of the release workflow now rewrites those addresses to the release's own downloads, checking every signature on the way. - SonarCloud reads tests as tests, and no longer analyses the dataset, the Android resources the Tauri CLI generated, or the SQLite migrations its PL/SQL analyser misread.
- SonarCloud analyses in CI. A workflow runs the scan on pull requests and
on
mainin place of SonarCloud's automatic analysis, with its settings insonar-project.properties(formerly.sonarcloud.properties). Commits that only carry data are not scanned. It runs both test suites with coverage first (pytest-cov for the pipeline,npm run test:coveragefor the web app), so SonarCloud shows what the tests reach instead of reading 0%. - TypeScript 7, Node 26 and Python 3.15. The type-check runs on
TypeScript 7, the native compiler:
tscfor the Worker, the tests and the admin's Node code, andsvelte-check --tsgofor the app. TypeScript 6 stays installed beside it, because svelte-check and SvelteKit's route types still need the JavaScript API that 7.0 does not have (the admin's svelte-check stays on it too). CI, the release builds and Cloudflare's build run Node 26, the LTS from 28 October; the pipeline, its tests and every data job run Python 3.15, which writes the shards byte for byte as 3.12 did - no re-download for anyone. The leaderboards and the dead-link log name their encoding, so they no longer crash on Windows under an older Python. A new test fails when one of the version pins drifts from the others, and Web CI now runs on workflow-only changes so it sees them. The full assessment is indocs/plan/16-toolchain-2026.md;docs/plan/17-typescript-7.1.mdprepares the move to TypeScript 7 alone once 7.1 is out.
🔐 Security
- certifi 2026.10.19 (GHSA-6q48-6c7w-9gw2), the CA bundle behind every
HTTPS request the pipeline makes. eMudhra's
emSign ECC Root CA - G3had intermediates whose private keys were compromised; certifi 2026.10.18 - the version pinned here - had already dropped every emSign root, and 2026.10.19 puts back the three that are not affected (G1, C1, ECC C3) while G3 stays out. So the pipeline was never trusting the bad root, but the advisory's range covered 2026.10.18 and pip-audit flagged it; the lock now names the fixed release, with its hashes.