Skip to content

sync-skill: audit ours.txt, and recognize a module-split release - #30

Merged
zpyoung merged 1 commit into
mainfrom
sync-lessons/v1.4.190
Aug 27, 2026
Merged

zpyoung merged 1 commit into
mainfrom
sync-lessons/v1.4.190

Conversation

@zpyoung

@zpyoung zpyoung commented Aug 27, 2026 •

Copy link
Copy Markdown
Owner
Files Added Deleted Net
Test 0 0 0 0
Prod 3 $\color{#1a7f37}{\Huge{\mathbf{+}}}$​115 $\color{#cf222e}{\Huge{\mathbf{−}}}$​2 $\color{#1a7f37}{\Huge{\mathbf{+}}}$​113

Four lessons from the v1.4.190 sync (#29). Three are
rules, edited into the steps that own them; one is a situation to recognize, appended to
sync-lessons.md (which was empty before this).

Nothing in ## Hard safety rules or the fix policy's "Never, on either gate" list is touched.


1. ours.txt loses upstream work silently — Step 6 (rule)

What happened. Nine of the 83 paths restored from ours.txt had upstream changes in the
v1.4.188→v1.4.190 range that the exception discarded outright. No conflict, no marker,
--verify-seams and --verify-residuals both clean, pnpm typecheck clean.

The expensive one was package.json. v1.4.190 adds an optional dependency:

+    "@vscode/windows-process-tree": "0.8.0",

pnpm-lock.yaml is upstream-owned and had already resolved to the tag, so restoring the fork's
package.json produced a lockfile mismatch. Step 8's pnpm install --frozen-lockfile fails there
with an error naming nothing about ownership.

config/reliability-gates.jsonc was the widest: 84 gates against the tag's 88, and structurally the
fork copy held nothing fork-specific at all — identical policy block, no fork-only gate, every
difference upstream's newer evidence. TerminalPane.tsx would have dropped the whole
TerminalProcessExitOverlay feature, which the tag's pty-connection-types deps already expect.

The edit. Step 6 gains a loop over ours.txt diffing each path across the tag range, the
git merge-file --diff3 recipe for replaying what the fork does not own, and the package.json
failure signature by name.

Evidence. Reproduce at 26d4bdafa5 (this run's ownership commit) with
git diff --numstat v1.4.188 v1.4.190 -- <path> over each line of ours.txt. Fixed in
8e1c5b50b0.

2. A ratcheted chokepoint fires on fork-only code — file-ownership.md (rule)

What happened. v1.4.190 routes every wsl.exe call onto runWslProcess (stablyai#15903, stablyai#15923) and
every child process onto runProcess/spawnProcess (stablyai#15746), each pinned by a boundary test that
scans all sources and asserts the offender list is empty. One fork file that neither the merge nor
the resolution touched failed five checks from that one cause — including an upstream test far
from the fork, because the fork's direct execFile bypassed that test's runWslProcess mock, so
the scan consumed the queued metadata frames and the parser rejected record kind F.

The reference already covers upstream tightening .oxlintrc.json, and explicitly scopes itself to
that file. This is the same shape arriving through tests, and the local Step 8 gate cannot see it
at all.

The edit. A new ## When upstream ratchets a chokepoint section: the tell (a boundary test whose
received array is a single fork-* path), read the detector before editing, migrate rather than
allowlist, take the idiom from the upstream sibling migrated in the same release, and the two
follow-on effects — upstream's mocks, and the fork's own.

Evidence. Job logs on run
33023192618; fixed in
b77bf17d2b, which turned all five green.

3. Reading a failing job's log — Step 10 (rule)

What happened. gh run view --log-failed --job <id> printed nothing and returned

run 33023192618 is still in progress; logs will be available when it is complete

A 16-way shard matrix across two Node versions is essentially never all-settled when the first
failure lands, so this is the normal case, not an edge one. gh api repos/zpyoung/orca/actions/jobs/<id>/logs returns the log immediately.

Also recorded: same shard number failing on both Node versions is deterministic and one alone is
the flake shape — a free first read before spending a re-run. And gh run rerun --failed refuses
with cannot be rerun; This workflow is already running until every job settles.

The edit. Step 10 gains the API command, the ANSI-stripping note, the shard-parity heuristic, and
the rerun timing.

4. A release that dissolves a monolith into modules — sync-lessons.md (situation)

What happened. eb1792985f (stablyai#15172) replaced src/main/ipc/pty.ts — 8,030 lines — with a
39-line barrel over 73 new modules. -X ours kept the fork's 8,112-line monolith and took all 73
siblings, so two complete implementations were live at once with separate module state. The merge
was conflict-free, --verify-seams passed 388/388 (the stale monolith still held every declared
line), and typecheck was clean.

The tell — --verify-residuals is the only check that sees it:

src/main/ipc/pty.ts: recorded +96/-1, measured +8111/-38

This is the third such release (v1.4.186, v1.4.187), which is why it is written as a situation rather
than left to be re-derived. The entry also distinguishes it from ordinary -X ours hunk damage — a
one- or two-line residual drift, repaired in place, which this same sync also hit twice
(tab-types.ts lost an import whose usage survived; web-session-tabs-sync.ts lost a new argument
at both fork call sites).

The right move is recorded as four steps, plus two things the manifest checks will not tell you:
cross-boundary consumers may need no edit at all, and config/max-lines-baseline.txt keeps a stale
inline entry for the retired monolith.

Evidence. Fixed in
6a79baa796; the manifest's seams and
residuals entries were repointed together in that commit.


Per Step 15 this PR is left open for review and merges nothing. The sync is already on main
(6e7ab8595d) and v1.4.191-rc.0.zy01 is dispatched.

Copilot AI lite review requested due to automatic review settings August 27, 2026 00:01

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

@zpyoung
zpyoung merged commit 341b026 into main Aug 27, 2026
48 checks passed
zpyoung pushed a commit that referenced this pull request Oct 19, 2026
…nce off the deprecated owner resolver (stablyai#30484)

The eight mirroring callers (#23-#30) read the owner through
runtimeEnvironmentIdWithLegacyFallbackUnlessTwin, so a twin is still not mirrored. Sort-order
persistence (#2) leaves out ids that two hosts publish, and ids whose only server proof is a repo
row (isUnprovenServerSshRoute), instead of writing them to this desktop's or a guessed server's
meta. Deprecated importers 40 -> 31.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants