Cryptograss:Moods/uploads-and-embeds/todo
From PickiPedia: A knowledge base of bluegrass, old time psychedelic jams, and other public domain music
The things this Mood is waiting on, and who has them. Kept here rather than in chat, because a queue that lives in scrollback is a queue nobody can see.
- ☐Run one read-only check to see why seven of the old drafts fail before they startOn delivery-kid: grep -o '"draft_type": *"[^"]*"' /mnt/storage-box/staging/drafts/*/draft.json. If the seven come back as anything but content (or missing), finalize treats them as not found and magent will write a loader fix so they publish from where they sit.
- ☐Publish the remaining unfinalized uploads, one at a time, tab open until it says pinnedSonic Offerings is next (it was encoding fine before its page closed). The Justin-uploaded ones are his to decide.
- ☐Deploy pickipedia, then paste the replacement Common.js — in that orderREADY since 01:22 on 9 October: blocksAgo, blueRailroadDate and instrumentForm all load on the live wiki. Copy the RAW text, not the rendered page: index.php?title=User:Magent/Common.js_after_131&action=raw (md5 6bb8dbdd207b85b09e86bbb76ebfe718). Needs interface-admin. Until the paste, the Instrument form alerts twice.
- ☐One delivery-kid --rebuild deploy once #138 and #152 are inRunning e4e00ed4 since 5 October, so everything through #137 is live and neither of those two is.
- ☐Decide how to close the RSS SSRF — whitelist, or network egress rulesConfirmed real — LocalSettings.php sets $wgRSSUrlWhitelist to "*", so any page can make the server fetch any URL. The feed hosts are not a fixed list in code; podcast_config.py reads them off {{PodcastFeed}} templates in Category:Podcasts, so a static whitelist breaks the Firehose whenever anyone adds a podcast. The alternative is denying the wiki container egress to link-local and private ranges. magent's view: the second, but it is a judgement call about how much friction to accept, so it is yours.
- ☐Answer the four questions about the two uploads that vanished on 24 SeptemberThe exact error text would settle it. Best guess is Caddy dropping POST bodies over HTTP/3, which is our own #79 from April, still unresolved.
- ☐Close the dead PRs in maybelle-config#132#69, #74 and #75 would restore the Coconut code deleted in #129. #86 now carries its own reasoning for closing.
- ☐Cancel the Coconut account and drop its key from the vaultThe only item still costing money. The code has been gone since #129.
- ☐Merge maybelle#170 — kept originals recorded with the release and counted by the auditSplit out of #152, which merged a minute before this commit landed on its branch. Touches delivery-kid, so it rides the same --rebuild deploy as #138.
- ☐Check the eleven Coconut remnants for anything worth keepingNone holds media, so keeping one means finding the original file, not keeping the pin. Six are Justin's "learning terrapin outside Denver venue", never published; the three of Sky's from 14-17 September are untitled.
- ☐Decide, per recording, whether to publish or delete the thirteen unfinalized uploads still in staging (4.2 GB)Mostly real recordings the Coconut failures left behind, possibly the only copies. Includes the learning-terrapin upload (21 MB), still on disk. Three look superseded (melodica, Station Inn, Pickipedia Dispatch). Six are Justin's — he decides on those.
- ☐Make finalize survive the browser tab closing, and run one long encode at a time8 October — two of the unpublished recordings were finalized at once. Both were encoding (a 19-minute video at 0.71x, ~27 minutes in all), both stalled at 21:46 the moment the second started, and both ended at 22:09:05, the same second, with no redeploy — what happens when the page closes, because finalize runs inside the page's connection. Long recordings need finalize to be a background job, and concurrent encodes queued.
- ☐Unpin the Coconut remnants once Sky has checked themThe list page is the record, so unpinning loses nothing. Needs a small change so the audit can unpin deliberately rather than anyone touching the server.
- ☐Move the zine styling with its stylesheet, emptying Common.js for goodThe last thing left in that page. Held back from #131 on purpose: it is bound to classes in MediaWiki:Common.css, and moving the script alone would split one feature across two homes.
- ☐Delete the phantom orphan-draft cleanup promised by delivery-kid's startup docstringmain.py says it runs an initial cleanup and a periodic task. Neither exists in the code. One-line fix, but it sent me looking for a scheduler that was never there.
- ☐Test whether Opus audio plays on SafariNever tested, and it applies to everything we have ever published, not just surround. If a video on the wiki is silent on an iPhone, this is the first thing to check.
Done (7)
- ☑Find out why the pickipedia deploy of #131 has not landedFound 9 October. #134 (the <todo> tag) made check-classes.php fail in Jenkins' verification gate, so no build since 5 October wrote a deploy marker. Fixed in pickipedia#135, which also runs the gate in GitHub CI; deployed 01:22.
- ☑Merge pickipedia#131 — three features out of MediaWiki:Common.js into the extensionRe-verified 8 October — Common.js has not been edited since 15 September, so the byte-identical comparison in the PR still holds. CI passed on the merge.
- ☑Merge maybelle#138 — surround audio bitrate5.1 was being published at the stereo rate, about 21k a channel. Updated 8 October per Justin — rungs below 720p get a stereo downmix; 1080p, 720p and audio-only keep every channel. Mono and stereo untouched, with a test pinning that. 113 pass.
- ☑Merge maybelle#152 — the audit stops calling our own records deletableCoconut tombstones reported as history, plus the CI job for the audit's tests. The kept-originals commit missed the merge by a minute and is now #170.
- ☑
- ☑Merge pickipedia#133 — stop a podcast feed deciding what the import bot writesClosed #101. Feed titles and descriptions were interpolated into template calls raw.
- ☑Publish "There's only one way to find out" from its stuck draftFinalized after re-running it with the tab left open. Proves the old drafts can publish, and that the earlier failures were the page closing.