Limit the memory (RAM) tests can use

[cehteh]

Not actually about the RAM prices ;)

I have the situation that some mutations creating memory leaks (replacing - with + etc). This should be expected and normal in mutation testing. Such test should be just aborted before the OOM kills the test. Currently I fix this with a reasonable 'ulimit' for the shell running cargo mutants. The unintentional consequence here is that the whole mutation/build process is memory constrained which means I have to give it more memory than necessary not to kill the build. Ideally one should be able to limit the memory of the actual test run only.

[sourcefrog]

Yep, that makes sense.

One complementary is to add an assertion after the size is calculated and before the memory is allocated, so the program will panic if it's too big: "the cache is never more than 1GB" or whatever.

[cehteh]

On Sat, 24 Jan 2026 09:39:44 -0800 Martin Pool ***@***.***> wrote: Yep, that makes sense. One complementary is to add an assertion after the size is calculated and before the memory is allocated, so the program will panic if it's too big: "the cache is never more than 1GB" or whatever.

I am thinking like sampling the memory usage for the baseline test run and using that for mutation runs (multiplied with a configureable factor, like *2.0 because sometimes test/failures may create huge debug data). But I explicitly left the implementation of this open. There are certainly enough other ways and considerations to do (portability).