Two CI checks currently guard test coverage for new code, and both have gaps:
The coverage job runs just coverage-ci to produce lcov.info and uploads it to Coveralls, which blocks a pull request only when total workspace coverage drops below a threshold. It does not identify which function lacks tests, and coverage from tested code in a large PR can mask an untested function.
rustqual's TQ-003 (run via qual diff) fails when a new production function is not called by any test. Because TQ-003 only traces direct calls in the AST, functions invoked through #[serde(default = "…")], clap, or derive macros are reported as untested (for example, default_true in crates/jp_config/src/conversation/label.rs). The 94 existing TQ-003 findings have not been triaged to separate genuine gaps from indirect callers, so TQ-003 is excluded from the burn-down (T-0yh5sy9) until this ticket lands.
When passed --coverage <lcov>, rustqual evaluates two LCOV-based rules instead of the static call graph:
TQ-004: a production function with zero covered lines.
TQ-005: a partially covered production function with uncovered lines.
Run through qual diff, both rules fail only on newly added functions and identify the untested function by name.
Where to run the coverage-based rules. qual diff needs lcov.info for both the head and base revisions, and the coverage job is the slowest step in CI whereas qual finishes in seconds. Either qual downloads the head lcov.info artifact from coverage and a cached artifact for base, or the coverage rules run as a step inside the coverage job.
Whether to enable both TQ-004 and TQ-005, or only TQ-004.
Whether this check replaces the Coveralls upload or runs alongside it.
How to handle local runs, since cargo_check does not generate coverage data and findings would only surface in CI.