Probabilistic (random) Mutation

[benedikt-mangold]

I would like to have a setting, where each suitable mutation is only applied with a probability of x (e.g. 50%) for one mutation-run.

Then, being able to configure N mutation-runs (e.g. N=100), getting the stats on

• average impact of a single mutation on the final score (independent of the others)
• quality score per function
  • average impact scores of mutations within a function

This allows to break down the quality statistics on a lower level (e.g. functions, lines of code) to point towards subparts that are tested particularly well / week

[boxed]

That seems like just sometimes running some mutants multiple times unnecessarily and missing some, while taking more time.

[benedikt-mangold]

From a statistical point of view, you can get insights on "how many mutants do you need to affect the results of a test". As an example - test is not failing, two mutants are present in the function. How do you know which one is actually responsible? By (randomly) eliminating one mutant (over the N runs), you can assign importances to mutants, depending on how many tests are affected solely due to a single mutant

[boxed]

two mutants are present in the function

That's never the case. Mutmut always tests exactly one mutant at a time.