On Fri, Nov 9, 2018 at 1:35 PM Theodore Y. Ts'o <tytso@xxxxxxx> wrote: > > How about creating a group "regress", and we can move them into > tests/regress for better categorization, for regression tests? "regress" group seems good, we'll send out a patch with that. > And for tests that attempt find new problems, we already have a group > "fuzzers"; and perhaps there should be a some standard guieline for > how long a single invocation of a fuzz test should run before > quitting. I'd say something in the 5 to 15 minute range? People can > always run the fuzz test multiple times, and if the fuzz test can save > test between runs, it should matter that it's broken up into multiple > test invocations. CrashMonkey doesn't do random tests for every invocation, so its behavior is a bit different from a fuzzer. But we can definitely modify CrashMonkey to randomly pick workloads in the target state space and run for a pre-defined amount of time.
