ipodderx-rs/TODO.md

# To do

## Errors in the log

From the production log and the feeds' stored errors on 2026-09-13. The Docker log only reached back
to 12:32 UTC, so the list comes from `feeds.last_error`: 57 of 1,059 feeds. None of it is ipx's
User-Agent; a browser gets the same answers.

- [ ] **Stop scanning an OPML's feeds once nobody subscribes to it.** 55 of the 57 are feeds from
  `davewiner` (lists.opml.org/davefeeds.xml). The list left `config.toml` about 14 hours before this
  was written, but its 922 feeds are still in the database and still scanned every hour, with
  auto-download on: `subscriptions()` adds every derived feed, and with no parent to copy from,
  `auto_download` falls back to `true` (`src/main.rs:937`). `docs/users.md` says a feed stops being
  scanned when its last subscriber leaves. Skip a derived feed whose parent is not in config, and
  when the last subscriber leaves an OPML, treat its feeds the way `sync_group` treats ones the list
  dropped: remove those with nothing downloaded, mark the rest orphaned. `remove_feed` and `ipx rm`
  both leave them behind today. (`src/main.rs`, `src/web.rs`)
- [ ] **Read feeds with a bare `&`.** kcpw has `<link>https://kcpw.org/?post_type=post&p=125715</link>`,
  and both feedland feeds have the same fault. Strict XML refuses them; browsers and other readers do
  not. When `feed::parse` fails, try once more with every `&` that does not start an entity written
  as `&amp;`. Nobody subscribes to these three now, but the next feed like them will fail the same
  way. (`src/feed.rs`)
- [ ] **Say what came back when it is not a feed.** Thirteen errors read "not RSS (the input did not
  begin with an rss tag) and not Atom (...)". Each one checked was a web page: the feed moved and its
  old URL redirects to the site, or the domain lapsed. Say "got a web page, not a feed", and when the
  page links a feed (`<link rel="alternate" type="application/rss+xml">`), name it. That link found
  the new feed for om.co, ms.now, Letters of Note, the Daily Dot, Hell Gate, The Frame Lab and
  Daily Kos. A `202` with an empty body (British Antarctic Survey) should read as "nothing yet", not
  as a parse failure. (`src/feed.rs`)
- [ ] **Show a publisher's error in the UI.** Today a failing feed shows its raw error in red only
  once you open it (`web/index.html:877`, `:960`); the OPML view marks a failing child "error"
  (`:988`), and the sidebar shows nothing. Mark a failing feed in the sidebar too, and say whose
  problem it is and what to do, in plain words: a 404 means the publisher took the feed down or moved
  it, 401 and 403 that the site refuses ipx, 402 that the feed needs a paid plan, a name that no
  longer resolves that the site is gone, and a web page instead of a feed that it moved (with the new
  URL when the page links one). Offer Unsubscribe, or Use the new address. Keep timeouts, 5xx and
  429 out of it, and flag a feed only once it has failed for a day: macmanx failed once and read
  fine an hour later. That needs the time a feed started failing, kept beside `last_error` and
  cleared when a check succeeds. (`web/index.html`, `src/db.rs`, `src/main.rs`)
- [ ] **Westenberg.** The only failing feed anyone subscribes to. The Hacker News stars list still
  gives `joanwestenberg.com/rss`, which is a 404; the feed is now `joanwestenberg.com/feed`.
  Nothing for ipx to fix; subscribe to the new URL directly.

The rest are the publishers' doing: 404 (24 feeds), 403 from sites that refuse anything but a
browser (5), no longer resolving or connecting (7), and one each of 400 (an invite-only Substack),
401 (a site gone private) and 402 (an expired rss.app trial). Once the first item lands, all of them
but Westenberg stop being scanned.

After these: `cargo test`, `node tests/page-smoke.js`, `npx playwright test`.