You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
# Miri tests take up a lot of memory, so only run 1 test at a time by default.
test-threads = 1
Previously, I'd have just agreed with the statement in the default config. But I've landed a number of PRs over the past months, culminating with rust-lang/rust#118336 which have all reduced the interpreter's memory growth over time. The interpreter is still not lightweight by any means; unit tests seem to consume no less than 250 MB and I rarely see them using more than 500 MB.
But I don't think suppressing parallelism is a clearly superior default anymore. I'd bet that most development environments can support available_parallelism() Miri jobs now, but I expect you want to be more cautious than that. Thoughts?
The text was updated successfully, but these errors were encountered:
Sure, sounds good to me. I don't remember actively making this decision -- I think I went along with something either @RalfJung or you said.
Do you want to be more cautious and increase the parallelism to (say) 4, or do you just want to remove the constraint entirely and let the settings from the default profile take over?
At some point I'd love to add arithmetic operations, so you can say something like min(system_memory() / 2GiB, num_cpus()), but that's a far-future thing.
Yeah that arithmetic would be ideal. If the only options are full parallelism or a fixed number, I'd prefer full parallelism. I suspect it will Just Work for nearly everyone, and I have little grasp of how many people are missing out on nextest parallelism because of this default, but I suspect that's the more common experience.
nextest/nextest-runner/default-config.toml
Lines 111 to 113 in ef47d01
Previously, I'd have just agreed with the statement in the default config. But I've landed a number of PRs over the past months, culminating with rust-lang/rust#118336 which have all reduced the interpreter's memory growth over time. The interpreter is still not lightweight by any means; unit tests seem to consume no less than 250 MB and I rarely see them using more than 500 MB.
But I don't think suppressing parallelism is a clearly superior default anymore. I'd bet that most development environments can support
available_parallelism()
Miri jobs now, but I expect you want to be more cautious than that. Thoughts?The text was updated successfully, but these errors were encountered: