This introduces a data race on mrng, which is not thread-safe. On close analysis, this RNG seems to be a copy-paste error from the fuzzer (maybe we should check to make sure the fuzzer also doesn't have this problem), and might be removable. If not, then the RNG is just used to seed a new RNG later on, so we could pre-generate the seeds for each subject in a map[string]int64 or something before parallelisation.
In the lifter, we have the following code:
This introduces a data race on
mrng
, which is not thread-safe. On close analysis, this RNG seems to be a copy-paste error from the fuzzer (maybe we should check to make sure the fuzzer also doesn't have this problem), and might be removable. If not, then the RNG is just used to seed a new RNG later on, so we could pre-generate the seeds for each subject in amap[string]int64
or something before parallelisation.