Today — 2.0.0 moved the gate to reconverge 0.4.0; its release is titled "reconverge 0.4.0, and the first ten minutes". The sites that install the analyzer moved with it: action/action.yml:54 (default: "0.4.0"), action/README.md:66, .github/workflows/prune.yml:20 (RECONVERGE_VERSION: "0.4.0"), and docs/LIMITATIONS.md:9 ("reconverge (v0.4.0)"). Three of the six sites the pin policy names did not, measured against 2.0.0:
$ grep -n 'reconverge' rust-toolchain.toml
9:# reconverge v0.1.11 (43780b58)
$ sed -n '163p' CONTRIBUTING.md
cuda-oxide 50d07314, reconverge 0.1.11 (installed from crates.io).
$ grep -n 'RECONVERGE_PIN' .github/workflows/pins.yml
20: RECONVERGE_PIN: v0.1.11
pins.yml:39-41 compares reconverge's latest tag against RECONVERGE_PIN and opens an issue on mismatch. #17 is that issue — "latest tag is v0.3.0, pin is v0.1.11" — filed against a repository whose gate has installed 0.4.0 since 2.0.0. The same workflow's issue text (pins.yml:56) is the list of sites a bump must touch: "rust-toolchain.toml, .github/workflows/pins.yml (env), .github/workflows/prune.yml (env), action/action.yml (input defaults), action/README.md, and CONTRIBUTING.md". main is one CI-only commit past v2.0.0 and changes none of these.
Why it is worth fixing — CONTRIBUTING §9 is what a contributor reads before touching a pin, and it states the wrong pin under the word "Current". rust-toolchain.toml's header is the lockstep record, and it records a set that no longer exists — the nightly and the cuda-oxide commit are right, the third member is not, and the comment above it says the three "move together or not at all". And the watcher's baseline is stale, so its weekly signal is noise: it reports movement away from 0.1.11 whatever 0.4.0 does, which is why #17 has been open since before the bump it is nominally about. The policy's own words are "every site, or not at all"; 2.0.0 did four of six.
Fix — set the three to v0.4.0 / 0.4.0 in one commit — the policy says a pin move is its own commit, and this is the tail of 2.0.0's. Then close #17 as superseded, or let the next scheduled run go quiet on its own. Worth adding while there: a step in pins.yml that asserts the six sites agree with each other before it asks upstream anything. That is the check that would have caught this, and it needs no network.
Done when — grep -rn 'reconverge' rust-toolchain.toml CONTRIBUTING.md .github/workflows/pins.yml shows 0.4.0 at every site; the pin watch's next run reports no reconverge drift, or drift measured from 0.4.0; the workflow fails when the six recorded pins disagree.
Today — 2.0.0 moved the gate to reconverge 0.4.0; its release is titled "reconverge 0.4.0, and the first ten minutes". The sites that install the analyzer moved with it:
action/action.yml:54(default: "0.4.0"),action/README.md:66,.github/workflows/prune.yml:20(RECONVERGE_VERSION: "0.4.0"), anddocs/LIMITATIONS.md:9("reconverge(v0.4.0)"). Three of the six sites the pin policy names did not, measured against 2.0.0:pins.yml:39-41compares reconverge's latest tag againstRECONVERGE_PINand opens an issue on mismatch. #17 is that issue — "latest tag isv0.3.0, pin isv0.1.11" — filed against a repository whose gate has installed 0.4.0 since 2.0.0. The same workflow's issue text (pins.yml:56) is the list of sites a bump must touch: "rust-toolchain.toml, .github/workflows/pins.yml (env), .github/workflows/prune.yml (env), action/action.yml (input defaults), action/README.md, and CONTRIBUTING.md".mainis one CI-only commit past v2.0.0 and changes none of these.Why it is worth fixing — CONTRIBUTING §9 is what a contributor reads before touching a pin, and it states the wrong pin under the word "Current".
rust-toolchain.toml's header is the lockstep record, and it records a set that no longer exists — the nightly and the cuda-oxide commit are right, the third member is not, and the comment above it says the three "move together or not at all". And the watcher's baseline is stale, so its weekly signal is noise: it reports movement away from 0.1.11 whatever 0.4.0 does, which is why #17 has been open since before the bump it is nominally about. The policy's own words are "every site, or not at all"; 2.0.0 did four of six.Fix — set the three to
v0.4.0/0.4.0in one commit — the policy says a pin move is its own commit, and this is the tail of 2.0.0's. Then close #17 as superseded, or let the next scheduled run go quiet on its own. Worth adding while there: a step inpins.ymlthat asserts the six sites agree with each other before it asks upstream anything. That is the check that would have caught this, and it needs no network.Done when —
grep -rn 'reconverge' rust-toolchain.toml CONTRIBUTING.md .github/workflows/pins.ymlshows 0.4.0 at every site; the pin watch's next run reports no reconverge drift, or drift measured from 0.4.0; the workflow fails when the six recorded pins disagree.