Minimization is the regression-suite management technique that permanently removes tests judged redundant against a criterion, typically coverage. Where several tests cover the same statements, one is kept and the rest are dropped (Yoo and Harman 2012)¹.

What it does

It caps suite growth by discarding coverage-redundant tests for good.

The danger is inherent in the criterion. Coverage redundancy is redundancy in the reachability link of the RIPR chain alone; two tests that execute the same lines can still differ in whether they infect and propagate state. A test that is redundant for coverage may be the sole detector of a real fault, and deleting it optimizes reachability while throwing that detector away.

Choosing the criterion

The same coverage profile supports opposite decisions. When read as gaps it argues for adding tests where code is unexercised; when read as redundancy it argues for removing tests that exercise only already-covered code. Minimization is that second reading of the same profile. The profile also marks what is not safe to drop: a test that is the sole cover of some statement is not redundant on the criterion, so a coverage-preserving minimizer must keep it. (Formally it is a set-cover problem: the smallest subset of tests whose union still covers every element.)

Mutation testing supplies a criterion stronger than coverage: it scores each test by the mutants it kills, and counts a test redundant only when every mutant it kills is also killed by another test. A kill exercises the whole RIPR chain, so mutation-redundancy accounts for the infection and propagation that coverage cannot see, and a test that uniquely kills a mutant is kept even where its lines are covered elsewhere. It costs far more to compute, which is why coverage stays the common criterion despite being the weaker one.

When to use, when not

Prefer selection (skip a test for this run) over permanent minimization (delete it for good), unless the suite's growth is genuinely unsustainable and the deleted tests are demonstrably redundant, not just coverage-redundant.

Don't minimize a suite to hit a size or coverage target. Deleting coverage-redundant tests to make a number move is Goodhart's-law gaming, the same failure mode that afflicts coverage targets; it buys a smaller suite at the risk of a lost fault detector.

Classification

• Quality dimensions: Maintainability (caps suite growth by permanently removing coverage-redundant tests) — permanently removes coverage-redundant tests; it prunes the suite, it does not run or verify it, so it sits off the code-verification axes.
• Area: Suites whose growth is genuinely unsustainable and whose redundant tests are demonstrably safe to delete.

Referenced by

• Mutation testing · Methods
• Regression-suite management · Methods
• The test suite as an object · Methods

References

---

1. Yoo, Shin, and Mark Harman. 2012. "Regression Testing Minimization, Selection and Prioritization: A Survey." Software Testing, Verification and Reliability 22 (2): 67–120. https://doi.org/10.1002/stvr.430. ↩