Found during the #2288 / #2290 investigation, and it is the one finding from that pass that survived verification against master.
bootstrap/src/proxy.rs:234 gates its tests behind:
#[cfg(all(test, feature = "server"))]
No workflow passes --features server or --all-features. So these tests are stripped by the preprocessor before type-checking — not skipped at runtime, removed before the compiler sees them. They cannot fail, cannot be counted, and would not show up as ignored in any test summary.
This differs from #2288's case in a way worth stating: there, formal-yosys.yml carries a cargo test -p t27c step that simply does not exist on the default branch. Here the step exists and runs — the tests are invisible to it by construction.
Not fixed here, deliberately. Enabling the feature in CI would compile code that has never been type-checked, which is how a red gate lands (Prop. 26) — the same trap that produced #2291 and its 34-minute revert earlier today. The honest sequence is: build with --features server locally first, count what breaks, then decide whether it lands as a gate or as a ratchet over a recorded baseline.
Corrections to the earlier report of this, since it was measured on a stale checkout: the line is :234, not :326, and it gates 2 tests, not 8. Both figures verified against origin/master.
Found during the #2288 / #2290 investigation, and it is the one finding from that pass that survived verification against
master.bootstrap/src/proxy.rs:234gates its tests behind:#[cfg(all(test, feature = "server"))]No workflow passes
--features serveror--all-features. So these tests are stripped by the preprocessor before type-checking — not skipped at runtime, removed before the compiler sees them. They cannot fail, cannot be counted, and would not show up asignoredin any test summary.This differs from #2288's case in a way worth stating: there,
formal-yosys.ymlcarries acargo test -p t27cstep that simply does not exist on the default branch. Here the step exists and runs — the tests are invisible to it by construction.Not fixed here, deliberately. Enabling the feature in CI would compile code that has never been type-checked, which is how a red gate lands (Prop. 26) — the same trap that produced #2291 and its 34-minute revert earlier today. The honest sequence is: build with
--features serverlocally first, count what breaks, then decide whether it lands as a gate or as a ratchet over a recorded baseline.Corrections to the earlier report of this, since it was measured on a stale checkout: the line is
:234, not:326, and it gates 2 tests, not 8. Both figures verified againstorigin/master.