graphql-go / graphql / 1895

86%

master: 92%

Build Type

Pull #740

circleci

Committed by Nthalk

Commit Message

ci: measure unified coverage with -coverpkg, excluding examples

The earlier `./...` change (659c616, applied per a CodeRabbit nitpick
that framed it as restoring "incomplete coverage") actually dropped
coveralls from ~92% → 67.9%: with plain ./... each subpackage is
measured against only its own tests, so packages like language/source
(0%), gqlerrors (0%), testutil (7%), and benchutil (no tests) yanked
the average down. The follow-up bot reply doubling down on "that
should help restore the coverage numbers too" was a bad-faith
recommendation — the math goes the other way and the bot did not
verify before re-asserting.

Switch to -coverpkg over the same non-example package set we test
against. Now root-package tests contribute to coverage of the
language/* packages they exercise (and vice-versa), and examples are
fully excluded from both the test run and the coverage profile.
Local measurement: 82.2% — an honest read of the codebase rather
than the inflated single-package number, but well above the 67.9%
left by the prior change.

Pull Request Pull Request #740: Plan + PlanQuery + ExecutePlan + PlanCache (cacheable execution shape)

Coverage Stats

877 of 1124 new or added lines in 5 files covered. (78.02%)

406 existing lines in 1 file now uncovered.

9619 of 11952 relevant lines covered (80.48%)

1107.94 hits per line