v3.2.7
Two fixes. Both were found by looking at what the project actually did rather
than at what it claimed, and one was found by running the artifact shipped
minutes earlier.
Fixed — the shipped artifact now tells the truth about itself
-
The shipped component announced
scry 0.0.0(#152, FEAT-083). Found by
running the SHIPPED v3.2.6 artifact minutes after cutting it — not 3.2.6, and
not the 3.2.4 that #133 had been about.The mechanism is worth recording because the previous fix caused it. FEAT-076
replaced a hand-maintained literal withenv!("CARGO_PKG_VERSION"), which is
correct under cargo. The shipped component is built by Bazel, and
rules_rustpopulatesCARGO_PKG_VERSIONfrom the target'sversion
attribute — which was absent, so it defaulted to0.0.0. A fix that is right
for one build system and silently wrong for the one that ships.Fixed in three layers, because the source-level assertion had become
tautological (SCRY_VERSION == env!("CARGO_PKG_VERSION")where the constant
IS that env): the Bazel target now carriesversion,claims.yamlpins the
literal, and CI byte-checks the built artifact withgrep -qaF— the-a
matters, since without it the check passed on both the fixed and the broken
wasm.
Fixed — CI
-
Clippy ran on 4 of 12 crates (#157, FEAT-086), the same defect as #141's
Test job one gate over. Two live clippy errors existed on main that the job
could not see. Both gates are now widened, and the coverage guard checks
both rather than one.The guard's first version scraped the workflow with
awkranges that chained
across the whole file, so its two lists came out identical — it could not
detect the drift it existed for, and printedFIRESunder mutation with the
wrong attribution. Replaced with a YAML-parsing checker shipping a
--self-testthat CI runs BEFORE the real check.Recorded plainly: the implementation merged as PR #159 titled
"FEAT-086 (#157)" while no such artifact existed. It was filed after the
fact (#160).rivet validatecannot catch that — it validates artifacts that
exist, not references to ones that do not. That gap is what #161's commit
traceability now closes.