Skip to content
This repository was archived by the owner on Nov 4, 2026. It is now read-only.

Commit f7850ee

Browse files
fix(devx): retire the dead CI half of the lockfile merge driver (#7018)
changeset-release.yml configured merge.pnpm-merge but performs no local merge, so the config could never fire. It was the last CI consumer. .gitattributes and CONTRIBUTING.md's contributor path are measured live and are kept; the guide section is re-pointed at what the mechanism now is: contributor-facing with zero CI consumers. The pin test's anti-vacuity weight moves rather than vanishing: configuring.length > 0 was proving both that the grep matched something and that the mechanism had a consumer. Job one moves to a control term that must hit the same population; job two moves to .gitattributes and CONTRIBUTING.md, the two files that are the mechanism today. The phantom direction becomes load-bearing. Part of #6436 Claude-Session: https://claude.ai/code/session_01GgDDqh6YnkXqsnVTCa7wHk Co-authored-by: Claude <noreply@anthropic.com>
1 parent 226ca38 commit f7850ee

8 files changed

Lines changed: 294 additions & 122 deletions

File tree

Lines changed: 25 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,25 @@
1+
---
2+
---
3+
4+
Removes the dead CI half of the `pnpm-lock.yaml` merge driver and re-points the mechanism at
5+
what it now is: contributor-facing, with zero CI consumers (objectui#6436, maintainer ruling of
6+
2026-08-27 adopting option B).
7+
8+
`changeset-release.yml` configured `merge.pnpm-merge` but performs no local merge — the version
9+
branch is updated by `reset --hard` plus a force-push inside `changesets/action`, and a
10+
force-push resolves nothing — so that configuration could never fire. It was the last CI
11+
consumer. `.gitattributes` and CONTRIBUTING.md's contributor path are kept: measured in a
12+
scratch repository with one variable changed, the attribute is what turns a conflicting lockfile
13+
merge into a regeneration instead of conflict markers left inside `pnpm-lock.yaml`.
14+
15+
The pin test's anti-vacuity weight moves rather than vanishing. `configuring.length > 0` was
16+
doing two jobs — proving the grep matched something, and proving the mechanism had a consumer.
17+
Zero is the correct count now, so job one moves to a control term that must hit the same
18+
population (`git config`, which `changelog.yml` runs) and job two moves to the two files that
19+
are the mechanism today: `.gitattributes` and CONTRIBUTING.md. The phantom direction becomes
20+
load-bearing, since it now asserts the page names no workflow at all.
21+
22+
Also re-points, for the second time, the worked-example citation in
23+
`scripts/check-pre-install-import-graph.mjs` that cited this config as "a quoted argument that
24+
must not be read as an install". No live in-repo example remains, so it is cited as formerly
25+
`changeset-release.yml`, anchored on the test case's name rather than a line number.

‎.gitattributes‎

Lines changed: 10 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,16 @@
11
# Git Attributes Configuration
22

33
# Use custom merge driver for pnpm-lock.yaml
4-
# This prevents manual merge conflicts in the lockfile by regenerating it
4+
# This prevents manual merge conflicts in the lockfile by regenerating it.
5+
#
6+
# ⭐ CONTRIBUTOR-FACING, with ZERO CI consumers (objectui#6436, ruled 2026-08-27).
7+
# An attribute only names a driver; it does not define one. The definition lives
8+
# in each contributor's local git config -- CONTRIBUTING.md, "Configure Git Merge
9+
# Driver for pnpm-lock.yaml" -- and no workflow defines it any more, because no
10+
# workflow performs a merge git carries out on the runner. Keep this line: with it
11+
# a conflicting lockfile merge is regenerated by pnpm; without it the same merge
12+
# leaves conflict markers INSIDE pnpm-lock.yaml, and a hand-resolved lockfile is a
13+
# high-blast-radius artifact that fails far from its cause.
514
pnpm-lock.yaml merge=pnpm-merge
615

716
# Keep Shadcn components as generated vendor code

‎.github/workflows/changeset-release.yml‎

Lines changed: 30 additions & 29 deletions
Original file line numberDiff line numberDiff line change
@@ -518,21 +518,32 @@ jobs:
518518
cache: 'pnpm'
519519
registry-url: 'https://registry.npmjs.org'
520520

521-
# ⚠️ MEASURED 2026-08-26 (objectui#6391): this job performs NO local merge,
522-
# so the driver configured below has no occasion to fire HERE. The step is
523-
# kept anyway — see the ⛔ at the bottom of this comment, which is the part
524-
# that matters.
521+
# ⛔ Do NOT add a lockfile merge-driver `git config` step back here.
522+
#
523+
# There used to be one, and it could never fire. Removed by objectui#6436
524+
# under the maintainer's ruling of 2026-08-27 ("dead in CI, live for
525+
# contributors"): this job performs NO local merge, so a driver configured
526+
# here had no occasion to run.
527+
#
528+
# ⭐ The MECHANISM is not retired — only this file's dead half is.
529+
# `.gitattributes` still routes `pnpm-lock.yaml` through `merge=pnpm-merge`,
530+
# and `CONTRIBUTING.md` still tells contributors to define that driver
531+
# locally before `git merge upstream/main`. Measured both ways in a scratch
532+
# repo, one variable changed: WITH the attribute the driver fires and the
533+
# lockfile is regenerated; WITHOUT it, nothing else altered, the identical
534+
# merge ends in `CONFLICT (content)` with conflict markers left inside
535+
# `pnpm-lock.yaml`. The mechanism is contributor-facing with zero CI
536+
# consumers, and this file is where the last CI consumer used to be.
525537
#
526538
# The sweep, so nobody re-derives it a fourth time (objectui#6358 and
527539
# objectui#6369 each re-derived it for a different workflow):
528540
#
529541
# Every `git` in this file, enumerated rather than grepped-for-absence:
530-
# the two `git config` lines below, `git status --porcelain` (x2),
531-
# `git checkout -- .` and `git clean -fdq` in "Restore the pre-version
532-
# tree". No `merge`, `rebase`, `pull`, `cherry-pick`, `am`, `apply` or
533-
# `revert` anywhere in the file — each of those zero-hits was taken with
534-
# a control term that DID hit the same file, because a zero-hit with no
535-
# control is not a reading.
542+
# `git status --porcelain` (x2), `git checkout -- .` and `git clean -fdq`
543+
# in "Restore the pre-version tree". No `merge`, `rebase`, `pull`,
544+
# `cherry-pick`, `am`, `apply` or `revert` anywhere in the file — each of
545+
# those zero-hits was taken with a control term that DID hit the same
546+
# file, because a zero-hit with no control is not a reading.
536547
#
537548
# `git checkout -- .` (below, in the restore step) takes tracked paths
538549
# from the INDEX to undo what `pnpm changeset:version` wrote. That is a
@@ -547,29 +558,19 @@ jobs:
547558
# and `config user.*`. The version-branch update is `prepareBranch`
548559
# (checkout, then `reset --hard github.context.sha`) followed by a commit
549560
# and a FORCE-PUSH. It never merges, rebases or three-ways anything, so on
550-
# this side too the driver has no occasion to fire. (Its `commitMode:
561+
# this side too the driver had no occasion to fire. (Its `commitMode:
551562
# github-api` path — not selected here; the default is `git-cli` — builds
552563
# the commit through the GitHub API instead, and the bundle reads no
553564
# `.gitattributes` and implements no merge driver at all.)
554565
#
555-
# ⛔ DO NOT delete this step "because it is dead" without a ruling. It is
556-
# dead only in the CI half. The `.gitattributes` line it depends on
557-
# (`pnpm-lock.yaml merge=pnpm-merge`) is NOT dead: `CONTRIBUTING.md` tells
558-
# contributors to configure this same driver locally and then to
559-
# `git merge upstream/main`, which is a real local merge on the attributed
560-
# path. Measured both ways in a scratch repo: with the attribute the driver
561-
# fires and the lockfile is regenerated; with the attribute removed and
562-
# nothing else changed, the same merge ends in `CONFLICT (content)` with
563-
# conflict markers in `pnpm-lock.yaml`. Removing the CI half forces
564-
# `ci-cd-pipeline-doc.test.ts` to demand the attribute's removal too, which
565-
# would break that live contributor path — so the two must be decided
566-
# together, not dropped as one more dead row. objectui#6391 carries it.
567-
- name: Configure Git merge driver for pnpm-lock.yaml
568-
run: |
569-
# Configure custom merge driver for pnpm-lock.yaml
570-
# This allows Git to automatically resolve lockfile conflicts by regenerating it
571-
git config merge.pnpm-merge.name "pnpm-lock.yaml merge driver"
572-
git config merge.pnpm-merge.driver "pnpm install --no-frozen-lockfile"
566+
# ⚠️ Re-adding the step reddens `ci-cd-pipeline-doc.test.ts`, which now
567+
# asserts that ZERO workflows configure the driver — in both directions. If
568+
# some future job really does merge locally on the runner, it needs the step
569+
# AND a row in the "Lockfile Merge Driver" table of
570+
# `content/docs/guide/ci-cd-pipeline.md`; the pin fails otherwise. Note that
571+
# pin greps workflow files for the config key, so even NAMING the key in a
572+
# comment here trips it — deliberately the safe direction (a false red, never
573+
# a false green), which is why this comment spells the attribute form.
573574

574575
- name: Turbo Cache
575576
uses: actions/cache@v6

‎CONTRIBUTING.md‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -64,6 +64,13 @@ git config merge.pnpm-merge.driver "pnpm install"
6464

6565
This configuration allows Git to automatically resolve conflicts in `pnpm-lock.yaml` by regenerating the lockfile using `pnpm install` instead of attempting a manual merge.
6666

67+
**This step is not optional bookkeeping — it is the only place the driver is defined.**
68+
`.gitattributes` routes `pnpm-lock.yaml` through a driver named `pnpm-merge`, but naming a
69+
driver does not define one, and no CI workflow defines it either (none of them performs a
70+
merge git carries out on the runner; see the "Lockfile Merge Driver" section of
71+
`content/docs/guide/ci-cd-pipeline.md`). Skip this and the `git merge upstream/main` below
72+
leaves conflict markers inside `pnpm-lock.yaml` instead of regenerating it.
73+
6774
### Create a Branch
6875

6976
```bash

0 commit comments

Comments
 (0)