timeout option giving up?

[Eh2406]

While trying to debug why rust-lang/cargo#5921 is timing out on CI, I tried adding a value for the timeout option. When I try it locally after running for 20 minutes, I get Test aborted: Giving up after 10000 child processes crashed with no other info on how to proceed. I know it is rued of me not to have a MCVE, but I don't know how to did in. How should I start debugging this?

[AltSysrq]

Whoops, there's code to generate error messages but nothing to actually print them except for the final test failure.

I'm planning to do a new release tomorrow, so I'll see about adding a "verbose" option or similar.

[Eh2406]

Thank you! I look forward to the new release!

Also can we reduce the number of tries to like 100 or 1000? 20 min was a long time to discover it was not doing anything.

[AltSysrq]

Hm, to get 10000 processes crashed in 20 minutes, you'd need an average process lifetime of 120ms, so if you didn't set a timeout below that, I'd expect some output from aborting anyway, unless you're getting segfaults or something.

Also can we reduce the number of tries to like 100 or 1000?

Yeah, this ties into something I was planning for #91, to allow capping the number of iterations in the shrinking process.

In the mean time, I'd suggest putting .no_shrink() on your top-level strategy so you can get at least some information.

[Eh2406]

It is possible that in flailing around trying to figure out what was going on I tried a very short time out. If that did not give enough time for the process to startup that would explain the problem. Maybe a warning about two short a timeout should go in the docs?

I was already using .no_shrink(), I am glad to here I was on the rite track.

[AltSysrq]

that did not give enough time for the process to startup

Ouch, I hadn't thought of that edge case before. I'll at least add a warning, but it may be possible to catch that happening directly.

[Eh2406]

Thanks for all the help! I am trying a longer timeout (3000) and we will see what the runtime is and output looks like. I am hopeful that it won't Giving up with a more reasonable number.

[AltSysrq]

0.8.7 has a few things that should help here.

• Config::verbose (or PROPTEST_VERBOSE) can be set to make proptest log what failures occur as they happen.

• See A no shrink setting. #91 for some added configuration options to control shrinking.

• There's now code to detect if the child process crashes or terminates before starting any tests so it doesn't go into a near infinite loop of spawning child processes.

[Eh2406]

Thanks!

Clear error message. ✔️ Better diagnostic tools. ✔️ Timeout working correctly. ✔️

One thing I did notes thow is that the verbose is only "real time" if the test is run with --nocapture. Which may be worth mentioning in the docs.