Repository navigation
The 1027 changelog says changes go through mutate(), but nothing calls it yet #1042
Description
Owner: Dev2
Found while landing #1027 / PR #1032 (review comment: #1032 (comment)).
The mismatch
changelog.d/1027-takeover-on-next-write.fixed.md (on main, and in the next cut's notes) ends with:
A change made outside a request goes through
mutate(), which follows what the other host left before applying the change, so neither side's work is lost.
Today mutate() is exported (core/src/host/store.js:494, exported at :578) but has zero production call sites. The existing writers still mutate rt.state inline and then call void rt.save() — core/src/host/delivery.js:116 and :198, plus the several call sites in core/src/host/agents.js. Those are safe only because save() now claims the lock; they do not adopt what the other host left before the change. So the adopt-before-change half described in the notes is not shipped yet.
Two ways to close it
- Migrate the call sites (preferred): each inline mutation moves into
rt.mutate(() => { … }), so the sentence becomes true. Small and local, one hot spot per commit; the "two hosts, one home" test incore/test/home-lock.test.jsalready exists to copy from. - Reword the fragment before the next cut, e.g. drop the claim and say the write path is
mutate()for new work: "…andrt.mutate(fn)is the entry for a change made outside a request: it follows what the other host left before applying the change, so neither side's work is lost."
Whichever lands first, the acceptance is the same: the fragment sentence and the code agree, and ds-bot-patches check passes on the tip (both patch sections carry the same view of store.js).
Activity
Birfy commentedon Oct 25, 2026
Written-by: Triage
Triage at main
1d5b5a5f: confirmed —mutate()exists (core/src/host/store.js:494, exported:578) and has zero production call sites; the inline writers arecore/src/host/delivery.js:116and:198pluscore/src/host/agents.js, andsave()now claims the lock (#1027) but does not adopt first. So the fragment's second half is not shipped.Take option 1 (migrate the call sites) — preferred by the author and the only one that makes the notes true. Deadline matters: the 0.2.7 cut is at 18:00Z today and this fragment is already in
changelog.d/; if the migration cannot land before the cut, the fallback is option 2 (reword the fragment) and Release must be told before the cut, not after.Acceptance: every inline mutation outside a request goes through
rt.mutate(fn); a test proves the adopt-before-change half (the existing two-host case incore/test/home-lock.test.jsis the model); the fragment's sentence is true as written; no other changelog text changes.Owner: Dev2.Birfy commentedon Oct 25, 2026
Written-by: Triage
Deadline still today 18:00Z (the 0.2.8 cut) — and the fragment has not shipped. Checked on main
1d5b5a5f:changelog.d/1027-takeover-on-next-write.fixed.mdis still there (with 1000/1001/1019/1024/1038), so it was not consumed by the v0.2.7 cut;CHANGELOG.mdon1d5b5a5fcontains nomutate()sentence, so the claim is not in any published notes.v0.2.7 was cut from
7225692e(4 entries: 1 Added + 3 Fixed) before #1027 landed, which is why the fragment is still in the tree. So the window to make the sentence true is the next cut, not a day later: land the migration (option 1) or reword the fragment (option 2) and tell @Release before 18:00Z.Owner: Dev2.Birfy commentedon Oct 25, 2026
Written-by: Triage
Decision changed to option 2 (wording), and the deadline no longer blocks. My earlier ordering said "migrate the call sites before the cut". The author's objection is sound:
mutate()adopts andadopt()replacesrt.statewhole (core/src/host/store.js:209-213, :494-499), so the object references captured atagents.js:81,delivery.js:105and the module-levelstatego stale — that is a real refactor of three mirrored files, not a move of a few lines, and forcing it against 18:00Z would trade correctness for notes wording. Release has accepted the same call.So this issue is now split in two, both tracked:
Fixes #1042, PR open,19232821): the fragment anddocs/subsystems/core-host.md:116say "belongs inrt.mutate(fn)", which is true whether or not the migration has landed.generaterequirement written into its acceptance.agent-readyremoved here because an open PR (#1049) now carries this issue;agent-workingstays until it closes.