Skip to content

docs(integration): record the branch retirement and audit the stale refs #666

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
RonenMars wants to merge 1 commit into
base: main
Choose a base branch
from
+133 −0
Open

docs(integration): record the branch retirement and audit the stale refs #666

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions docs/followups/repo-health/06-integration-branch-decision.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
# 06 — Retire the long-lived integration branch

> **Resolved 2026-08-12 — retired.** Every open PR was landed on `main` individually (20 merged, 4 closed unmerged) and the integration branches were deleted. #575 and #580 were closed rather than conflict-resolved, their content already being on `main` via #578/#579. The decision, the deleted refs, and the audit that made deletion safe are in [`integration-branch-retirement-2026-08-12.md`](../../integration-branch-retirement-2026-08-12.md). The text below is the state of play that produced the decision — kept for the reasoning, not as open work.

**This is a decision, not a task. Do not execute it.** Gather anything missing, present the options, stop.

## State of play
Expand Down
131 changes: 131 additions & 0 deletions docs/integration-branch-retirement-2026-08-12.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,131 @@
# Integration branches — retirement and stale-ref audit, 2026-08-12

Closes the decision left open by [`followups/repo-health/06-integration-branch-decision.md`](./followups/repo-health/06-integration-branch-decision.md), and records what remains.

Companion to [`integration-to-main-2026-08-09.md`](./integration-to-main-2026-08-09.md), which logs the merge run itself.

## Decision: retired

The long-lived integration branch is gone. Every open PR was landed on `main` individually instead — 20 merged, 4 closed unmerged — and `main` was shipped from that state as build 198 / versionCode 50.

Brief 06's two open items are both closed:

| Item | Outcome |
| --- | --- |
| Record a decision — retire, or keep under two rules | **Retire.** Executed 2026-08-09 → 2026-08-11 |
| #575 and #580 need a path that is not manual conflict resolution | **Closed unmerged.** Their content was already on `main` via #578/#579, verified byte-identical against their own commits' file sets — no conflict to resolve |

The branch that replaced it, `integration/fresh-2026-08-09` (`e4f6056c`), was itself retired once `main` held its content. That is the intended lifecycle: an integration branch is a staging area with an expiry, not a parallel trunk.

## What was deleted

| Ref | SHA | Scope |
| --- | --- | --- |
| `integration/fresh-2026-08-09` | `e4f6056c` | remote + local |
| `integration/open-prs-291-544-…-568-569` (the branch brief 06 is about) | `0a4dd2d5` | remote + local |
| `integration/open-prs-291-544-…-568` | `d33ef5f4` | remote |
| `integration/2026-08-08-1740-prs-544-…-560` | `07376543` | remote |
| `integration/open-prs` | `9580ddfe` | remote |
| `integration/open-prs-544-551-553-554-556` | `5079c28d` | local only |

Two worktrees holding them were removed with them.

**Retained deliberately:**

```
origin/backup/int-mobile-2026-08-09-fresh 0a4dd2d5 # identical to the old INT
origin/backup/integration-open-prs-2026-08-09 60aef087 # the pre-#577-revert state
```

Deleting the old INT cost nothing because `backup/int-mobile-2026-08-09-fresh` points at the same commit. The second backup is genuinely distinct — it predates the #577 jest/TypeScript revert.

## The audit that made deletion safe

The question for each ref is not "is it merged" — none of these are ancestors of `main`, because every PR that fed them was squash-merged under a new SHA. The question is **does it hold a file that `main` has never had.**

```bash
# for each candidate ref: files present there and absent from main, ignoring docs
git diff --diff-filter=A --name-only origin/main "$REF" | grep -v '^docs/' |
while read -r f; do
# empty result = main never deleted it = main never had it
[ -z "$(git log --diff-filter=D -1 --format=%h origin/main -- "$f")" ] && echo "NEVER on main: $f"
done
```

Across all 16 surviving integration-named refs: **zero files were ever unlanded.**

| Ref | SHA | Last commit | Non-doc files not on `main` | Never on `main` | Remote |
| --- | --- | --- | --- | --- | --- |
| `27-07-2026.09-47-integration` | `28b13601` | 2026-07-27 | 17 | **0** | local only |
| `archive/26-07-2026.18-44-integration` | `eb330e71` | 2026-07-31 | 17 | **0** | yes |
| `chore/cherry-pick-sentry-fix-onto-integration-prep` | `57a3852a` | 2026-07-31 | 17 | **0** | local only |
| `feat/in-chat-search-integration` | `0ce45291` | 2026-07-23 | 19 | **0** | local only |
| `feat/live-external-sessions-integration` | `e87619c6` | 2026-07-23 | 18 | **0** | yes |
| `fix/diff-viewer-full-width-lines-integration` | `1f255696` | 2026-07-26 | 17 | **0** | local only |
| `fix/integration-branch-ci-failures` | `422709d5` | 2026-07-25 | 19 | **0** | yes |
| `integration-dev-merge-sim` | `33ecfc31` | 2026-07-23 | 19 | **0** | local only |
| `integration-dev-merge-sim-2` | `33ecfc31` | 2026-07-23 | 19 | **0** | local only |
| `integration-dev-merge-sim-3` | `33ecfc31` | 2026-07-23 | 19 | **0** | local only |
| `integration-dev/v1.0.0-2026-07-22` | `5502eb34` | 2026-07-23 | 19 | **0** | local only |
| `integration-dev/v1.0.0-df91938-2026-07-22` | `c9a78cd4` | 2026-07-22 | 18 | **0** | local only |
| `integration-merge-354-355-376` | `597f0293` | 2026-07-25 | 19 | **0** | local only |
| `integration-merge-354-355-376-v2` | `a2e94002` | 2026-07-26 | 17 | **0** | yes |
| `integration-merge-354-355-376-v2-rebased-main` | `7bfd942d` | 2026-07-26 | 17 | **0** | local only |
| `land/integration-prep` | `28c0c374` | 2026-08-01 | 16 | **0** | yes |

`integration-dev-merge-sim`, `-2` and `-3` are three names for one commit.

### What those 16–19 files actually are

The same set every time — content `main` **deliberately removed**:

| Content | Removed from `main` by |
| --- | --- |
| ~16 Maestro flows (`e2e/shots-0*.yaml`, `live-chat-tab.yaml`, `keyboard_session_input.yaml`, `servers_status_dots_menu.yaml`, …) | **#531** `test(e2e): delete the maestro flows that nothing runs` — a PR whose entire purpose was deleting them |
| further `e2e/*.yaml` including `codex_parity_real.yaml` | `243d5be4` |
| `crash-log.txt`, `log-mobile.txt` | #458 — stray logs |
| `KICKOFF-landing-runbook.md` | **#474** `docs: retire the landing kick-off…` |

So these branches are pre-cleanup snapshots. **Opening a PR from any of them would revert #531, #474 and #458** under a title that reads like recovery. That is the specific reason none of them should be "merged back".

**Security consequence.** `e2e/live-chat-tab.yaml` carried an API key as an `-e "API_KEY=…"` example. #531 removed it from `main` by deleting the flow, but the stale branches predate that deletion and still carry it — verified directly:

```
$ for b in land/integration-prep integration-merge-354-355-376-v2 archive/26-07-2026.18-44-integration; do
git show "${b}:e2e/live-chat-tab.yaml" | grep -c API_KEY; done
3
3
3
```

Retiring these refs reduces how many places hold that blob, but **it does not remediate it** — the key is in history that any existing clone already has. Rotation is the fix; deletion is hygiene.

### `integration-merge-android-r8` — a misnamed commit

A detached worktree sits at `f2242bc4`, titled `fix(android): pin expo-modules-core for R8 ColorCompat`. It contains **no Android code**. Its entire diff is four `SPEC CHECKSUM` lines in `ios/Podfile.lock`:

```
ExpoModulesCore 71242aac → a4d8d4e5
ExpoModulesJSI ef7f197a → b20fefa4
ExpoModulesWorklets …
ExpoModulesWorkletsAdapter …
```

That is the path-dependent pod noise documented in `CLAUDE.md`, not a fix. Anyone searching for that R8 fix on the strength of the commit subject will not find it, because it was never written here.

## Recommendation

**Drop all 16, local and remote. Do not open a PR from any of them.**

Two need an owner's nod rather than an audit result, and are held back pending that:

- `archive/26-07-2026.18-44-integration` — explicitly named *archive*, so someone may intend it as a keepsake. It is empty of unlanded content either way.
- `feat/in-chat-search-integration`, `feat/live-external-sessions-integration` — **feature** branches that merely contain the word "integration". They belong to feature work, not to this rollup effort, and should be judged on that basis.

## The method note worth keeping

Two checks in this audit return a confident wrong answer rather than an error, and both were hit here:

**A branch-name glob is a filter, not an inventory.** `git branch -a --list '*integration/*'` reported **zero** remaining branches, and that was reported as fact. It matches only refs with a literal `integration/` — every hyphenated name (`integration-merge-…`, `integration-dev-…`, `land/integration-prep`) was invisible to it. The true count was 23 refs. Use `git branch -a | grep -i <word>` when the question is "what exists", and reserve globs for when the pattern *is* the question.

**"Not an ancestor of `main`" does not mean "unlanded".** Squash-merging gives the landed content a new SHA, so ancestry calls every merged PR unlanded. The file-level "was it ever deleted from `main`" test above is what actually answers it. This is the same trap recorded in [`troubleshooting.md`](./troubleshooting.md) → "CI signals", where four tools in one day each returned a plausible wrong answer instead of failing.
Loading