fix(firms): keep the IPv4 retry from crashing the server; guard the call sites

- fetchIpv4 builds its Response inside the https response callback, outside
  the promise. A 204, 205 or 304, a status above 599 or an invalid reason
  phrase made the Response constructor throw there, an uncaught exception
  that stops the dev server. Null-body statuses now get a null body and
  anything else that cannot become a Response rejects, as the Radio Browser
  transport does.
- A non-OK FIRMS or mapkey-status response cancels its body so the IPv4
  socket closes at once instead of at the deadline.
- Tests: the crash cases (fail on the old code), a source check that every
  upstream request goes through fetchFirms and that fetchSource's body read
  shares its deadline (fails if either call reverts to fetch), a real-fetch
  contract test for the address-selection error shape, and an explicit skip
  in the IPv4-only socket test where the host cannot tell the families apart.
- Changelog entry moves to changelog.d/1078-firms-ipv4-fallback.md per
  CONTRIBUTING; CURRENT-STATE says which timeouts are retried and that the
  retry does not follow redirects.
This commit is contained in:
Said Turksever committed 2026-10-30 14:50:27 +00:00
1 parent bae3433b24
commit df01964171
4 files changed
+161 -22

No files matched your search

+4
View File
@@ -0,0 +1,4 @@
- FIRMS fire data loads on hosts where IPv6 is unreachable and IPv4 is slow to
connect: a request whose every address fails to connect retries once over
IPv4, and FIRMS warnings name the underlying connection errors instead of
only `fetch failed` (jeffblakely, #1078).