oficina / docs Public

Give the browser a way out of a corrupt snapshot (Task 7)

The server serves a snapshot without ever calling into the CRDT core,
so it cannot detect one corrupted on disk after it was written and
verified. Y.readUpdate throwing is the browser's own detection of
that, and until now there was no way out: the resync controller would
refetch the same corrupt bytes forever, leaving the document
permanently unloadable in the editor while it went on rendering fine
server-side.

live-updates.mjs's fetchSince now takes a second argument carrying a
raw flag; drain() retries the same cursor once with raw:true when
applying fetched bytes throws, and only advances the cursor once
something actually applied — a second throw reports through onError
and leaves the cursor alone. editor-entry.js wires the same fallback
into fetchSince and into attemptResync's own initial whole-log fetch,
which runs before any subscriber exists and would otherwise stop the
editor mounting at all. The retry emits the literal query string
raw=1, matching handlers.go's raw=1 escape hatch exactly.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>