Repository navigation
chore(deps): record the built_collection 6.0.0 hold decision - #163
Conversation
The issue blamed code_builder 4.13.0 and the build_daemon chain, but those already allow built_collection >=5.1.1 <7.0.0 — waiting on code_builder upstream would never release the hold. The sole blocker is built_value 8.13.0, which pins built_collection ^5.0.0 and is reached here via code_builder/build_runner/build_daemon. zorphy declares no direct built_collection constraint, so there is nothing to widen, and pinning 6.x would only make the graph unsatisfiable. Record the decision in-tree so the next release pass does not re-chase code_builder, with the upstream request tracked at google/built_value.dart#1405. Metadata only: no lib/, generated-output or lockfile change.
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
There was a problem hiding this comment.
PR review — #163
Verdict: approve as-is. One file, seven added lines, zero behavior change — a comment block recording the built_collection 6.0.0 hold decision.
No actionable issues
I re-derived every claim in this PR independently and found no defects to fix.
Claims I verified
- zorphy declares no direct
built_collection/built_valueconstraint —greponzorphy/pubspec.yamlfinds only the new comment lines. The existingbuild: ">=4.0.4 <6.0.0"note (lines 12-13) is indeed the same voice this comment matches. code_builder4.13.0 /build_runner2.17.1 /build_daemon4.1.7 allow 6.0.0 — read from the resolved pub cache:>=5.1.1 <7.0.0,>=5.1.1 <7.0.0,>=5.0.0 <7.0.0. So waiting oncode_builderupstream would genuinely never release the hold.built_valueis the sole blocker —built_value8.13.0 (latest) pinsbuilt_collection: ^5.0.0, and the last six published versions (8.12.3 → 8.13.0) all do.- 6.0.0 is a real breaking major — the
dart-lang/buildchangelog confirmsBuiltList.asList,BuiltMap.asMap,BuiltListMultimap.asMapandBuiltSetMultimap.asMapnow return unmodifiable views instead of copies. - The quoted solver error reproduces byte-for-byte —
dart pub add "built_collection:^6.0.0" --dry-runfails with exactly the two lines in the PR body (build_runner >=2.8.0 requires built_collection ^5.0.0→version solving failed). - The release gate is unaffected —
test/release/dependency_constraints_unchanged_test.dartreads onlydependencies/dev_dependenciesscalar entries throughpackage:yaml; a YAML comment is not a node, soexpectedZorphyDepsstill compares equal. Ran it:+1: All tests passed! pubspec.lockis unchanged —dart pub getleft the tracked lockfile byte-identical (git diff --numstatempty), anddart pub outdatedstill reportsbuilt_collection 5.1.2 → 6.0.0as Resolvable*5.1.2, i.e. the hold is unchanged and now documented.google/built_value.dart#1405exists, is open, and is titled "built_collection 6.0.0 is blocked by the ^5.0.0 constraint" — the tracking link is correct.
Non-blocking observations (no change requested)
- 🔵 The comment says "NOT by this constraint", but the preceding
code_builder: ^4.10.0line is what it sits above. A reader skimming line 27 first could misread which pin is meant. The wording is already unambiguous in the full sentence, so this is cosmetic only. - 🔵
5.1.2is hardcoded in the comment while the resolution moves — that is a snapshot, not a defect, but it will read as stale oncebuilt_valuewidens. Issue #158 stays open (state:OPEN), and itsActionsection still offers "wait forcode_builderupstream" — which this PR proves is the wrong branch. Worth a one-line correction on #158 so a future reader does not re-derive the false premise.
Verification I ran (macOS, Dart 3.13.x)
dart pub get→Got dependencies!, zero lockfile driftdart test -j 2 test/release/dependency_constraints_unchanged_test.dart→+1: All tests passed!dart test -j 2 test/release/→+11: All tests passed!dart pub add "built_collection:^6.0.0" --dry-run→ reproduces the documented failure
All CI checks on the head are green (analyze, test, format, build, prepare, analyzer_compat ×2, build_runner_smoke, migrator).
| # `built_collection` 5.1.2 -> 6.0.0 is held, and NOT by this constraint: | ||
| # code_builder 4.13.0 and build_daemon 4.1.7 already allow >=5.1.1 <7.0.0, so | ||
| # waiting on code_builder upstream would never release it. The sole blocker is | ||
| # built_value 8.13.0 (`built_collection: ^5.0.0`), reached here via | ||
| # code_builder/build_runner/build_daemon. Decision (issue #158): wait for | ||
| # built_value to widen rather than pin 6.x, which would make the graph | ||
| # unsatisfiable. Tracking: google/built_value.dart#1405. |
There was a problem hiding this comment.
🔵📐 Cosmetic only — no change requested. The sentence "built_collection 5.1.2 -> 6.0.0 is held, and NOT by this constraint" sits directly above code_builder: ^4.10.0 (line 27), so a reader who lands on the pin first can read "this constraint" as that one. The full sentence makes the intended meaning clear (no direct built_collection constraint exists anywhere in this file), so I would not block on it. Verified independently: built_value 8.13.0 is indeed the sole blocker, code_builder 4.13.0 / build_runner 2.17.1 / build_daemon 4.1.7 all allow 6.0.0, and the quoted solver error reproduces verbatim. The 6.0.0 breaking-change note also matches the published changelog.
Walkthrough — PR review 163 (arrrrny/zorphy)Head reviewed: Scope: 1 file, +7 / -0 — a comment block above Verdict: No actionable issues — approve as-is. How I verified itRather than take the comment's word for it, I re-derived the whole dependency story from the resolved pub cache and the published changelogs:
Tests run on this head (macOS, Dart 3.13.x, CI: all checks green on the head (analyze, test, format, build, prepare, analyzer_compat ×2, build_runner_smoke, migrator). This bot's own review check is the one pending entry. Non-blocking notes (no change requested)
|
Merge taskmerged as
Note: the merge itself hit a stale GitHub-side merge lock — REST Separate pre-existing base problem, unrelated to this PR's one-comment diff and filed for the pool: |
What
Issue #158 asks whether to wait for
code_builderupstream to widen itsbuilt_collectionconstraint, or pursue an upgrade path. Neither premiseholds — and this PR records the corrected decision in-tree.
The issue's blocker list is wrong. Read from the resolved pub cache, the latest
releases already allow
built_collection6.0.0:built_collectioncode_builder>=5.1.1 <7.0.0— allows 6.0.0build_runner>=5.1.1 <7.0.0— allows 6.0.0build_daemon>=5.0.0 <7.0.0— allows 6.0.0built_value^5.0.0— blocks 6.0.0The sole blocker is
built_value, whose every published version (8.10.1 …8.13.0) pins
built_collection: ^5.0.0. Waiting oncode_builderupstreamwould never release the hold. The solver says so directly:
There is no upgrade path:
zorphydeclares no directbuilt_collection(or
built_value) constraint, so there is nothing to widen here, and pinning6.x would only make the graph unsatisfiable. 6.0.0 is also a real breaking
major —
BuiltList.asList/BuiltMap.asMapnow return unmodifiable viewsinstead of copies.
Decision: wait, and carry the unblock upstream at
google/built_value.dart#1405 rather than re-investigating
code_builderonevery future release pass.
Change
A comment above
code_builder: ^4.10.0inzorphy/pubspec.yamlrecording thereal blocker and the decision. That file already documents constraint reasoning
in the same voice (see the
build: ">=4.0.4 <6.0.0"note above it), so theheld-upgrade note belongs there.
No
lib/, generated-output, or lockfile change —zorphy/pubspec.yamlis theonly file touched, and
pubspec.lockis unchanged.Why not change a constraint
test/release/dependency_constraints_unchanged_test.dartpins every declaredconstraint exactly (
expect(zorphyDeps, expectedZorphyDeps)), so adding abuilt_collectionconstraint would fail the release gate. A comment is notparsed, so the gate is unaffected — verified below.
Test evidence
cd zorphy && dart test -j 2—+374: All tests passed!(after
cd example && dart run build_runner build --delete-conflicting-outputs, which is the fixture step CI'spreparejobperforms; without it 15 tests fail in
setUpAllon missing fixtures)dart pub get—Got dependencies!, zero lockfile drift(
git diff --stat -- '*pubspec.lock'empty)dart analyze lib—No issues found!dart test test/release/dependency_constraints_unchanged_test.dart—+1: All tests passed!cd zorphy && dart test -j 2—+374: All tests passed!Note:
dart format --output=none --set-exit-if-changed .inzorphyreports18 changed files, and
dart analyzereports 315 issues underexample/. Bothare pre-existing on
masterwith the local Dart 3.13.7 SDK and areunchanged by this PR (18 before, 18 after;
dart analyze libis the cleanscope). Not touched here.
Closes #158
Summary by CodeRabbit