code-coverage – Page 5 – Türker Yıldırım

A golden master, for the report nobody can specify

NoteOctober 24, 20202 min read

A legacy report with no specification and forty branches cannot be unit tested, and it can be pinned: record what it produces today and fail when that changes.…

Mutation testing finds the assertion you did not write

NoteSeptember 16, 20201 min read

Ninety-four percent coverage and a suite that passes with the comparison inverted, because coverage measures execution rather than verification. The three numbers say different things: MSI is over…

Running mutation testing on the diff only

NoteAugust 6, 20201 min read

A full mutation run on a large codebase is hours, which means it runs nightly and nobody reads it. Running it on the changed lines takes minutes and…

Coverage of changed lines, not of the codebase

NoteJuly 28, 20201 min read

Total coverage moves so slowly that it says nothing about the change in front of you, and the question a reviewer has is whether the new code is…

Mutation testing, and the tests that assert nothing

Deep diveJune 18, 202010 min read

94% coverage, and a suite that passes with the logic inverted. Coverage measures execution rather than verification, and there is a tool that measures the difference.

Equivalent mutants, and why the score is never 100

NoteMarch 17, 20201 min read

Some mutations produce code that behaves identically to the original, so no test can possibly kill them — and they are indistinguishable from real findings without reading each…

Property-based testing, and the input you would not have written

NoteMarch 4, 20202 min read

Example-based tests check the cases somebody thought of, and the bugs are in the cases nobody thought of — which is a machine’s job. Finding the property is…

PHPUnit 9 and the assertions that finally went

Deep diveFebruary 25, 20209 min read

What 8 deprecated, 9 removes, and the deadline was a year. Rector does the renames; the one it cannot decide is the one that changed behaviour.

Coverage configuration moved, and the report went silent

NoteJanuary 16, 20202 min read

9.3 moved the coverage settings out of the filter element into their own, and the old form is ignored rather than rejected. The failure is silent and reports…

A baseline is what makes static analysis adoptable

Deep diveDecember 10, 201914 min read

0.12 shipped in December and the ignoreErrors list can be deleted. A generated debt register, a ratchet that only shrinks, and a level to stop at.