fix: keep dependency docs in sync - #50
Conversation
Signed-off-by: nightcityblade <nightcityblade@gmail.com>
1e42135 to
114a3e4
Compare
The 2026-09-08 update entry described the 0.9 → 0.10.1 move, which is what happened on that date; rewriting it to say 0.11 made a dated record claim something that was not true when it was written, and left the bullet below it still explaining that "0.10 records styles by default". termlens 0.11 landed here on 2026-09-11 in 2b12da6. So the 2026-09-08 entry keeps its history, and 0.11 gets its own dated entry: the stability-candidate requirement, the one breaking change (`Screen::unsupported()` returning a view, `unsupported_overflow()` folding into it) and what that bought both PTY suites, plus the widened `make skill-version`. ADR-0012 makes these records the audit trail, so a dated entry is evidence rather than a description of the present. SPEC.md and dependencies.md are current-state documents and are correctly rewritten in place. Also drops a dead line from the new guard: the first grep built `termlens/0/11` from `tr . /` and could never match `termlens/0.11`, so the regex beside it was doing all the work. One pattern now, with the spellings it covers named. Signed-off-by: Vyncint Ng <115854244+vyncint@users.noreply.github.com>
vyncint
left a comment
vyncint
left a comment
There was a problem hiding this comment.
Thank you. This is the right shape for the issue: the documents are corrected and a check now holds them to the manifest, so the next bump cannot leave them behind the way this one did.
I verified the guard rather than trusting it. Making docs/SPEC.md stale again (MSRV back to 1.88, the docs.rs link back to 0.10.1) fails it with docs/SPEC.md does not name rust-version 1.94.1, and restoring the file passes. All five SkillVersionTests pass, including your new test_stale_documented_dependency_version_fails. The remaining MSRV 1.88 mentions in dependencies.md are upstream crates' own declared MSRVs, which are correct as they stand.
I pushed one commit.
ADR-0011. The update entry is dated 2026-09-08 and described the 0.9 → 0.10.1 move, which is what actually happened then; termlens 0.11 landed here on 2026-09-11 in 2b12da6. Rewriting that dated entry to say 0.11 made it claim something untrue for its own date, and left the bullet below still explaining that "0.10 records styles by default". Under ADR-0012 these records are the audit trail, so a dated entry is evidence rather than a description of the present. The 2026-09-08 entry keeps its history and 0.11 gets its own dated entry covering the stability-candidate requirement, the Screen::unsupported() change and the widened make skill-version.
That distinction is on me rather than you: issue #37 listed ADR-0011 next to SPEC.md and dependencies.md without saying that the first is a historical record and the other two are current-state documents. You followed the issue as written. SPEC.md and dependencies.md are correctly rewritten in place.
One dead line in the guard. The first grep built termlens/0/11 via tr . /, which can never match termlens/0.11, so the regex beside it was doing all the work. Collapsed to the one pattern, with the spellings it covers named in a comment.
Merging. This closes a real gap, and the test you added is the part that keeps it closed.
2 participants
What and why
Bring SPEC.md, dependencies.md, and ADR-0011 in line with the manifest's termlens 0.11 and Rust 1.94.1 requirements. Extend the existing skill-version check so future dependency bumps also verify both normative dependency documents.
Fixes #37.
How it was verified
The five focused SkillVersionTests pass, including a regression proving stale documentation fails.
make skill-versionandgit diff --checkpass.make gateis green locallyChecklist
STATUS.mdupdated if a component's verification status changed (N/A)CHANGELOG.mdUnreleased updated for user-visible changes (N/A)unsafeblocks carry a// SAFETY:comment (N/A)datafusion-*sub-crate; dependency chain still coherent (make coherence)docs/decisions/(N/A)Co-Authored-Bynaming an assistant,no "Generated with" watermark. You are the author of record; CI checks this.