On Thursday, September 25, 2014 2:01:03 PM UTC-4, Jakob Kummerow wrote: > > That depends entirely on the kind of performance delta we're talking about. > > If a particular testcase went from actually hanging (endless loop, or > running for hours) to finishing quickly (say, in 1 millisecond), then by > all means land that testcase. (You don't have to detect the hang yourself, > the test driver already has a timeout for each test.) >
Okay. That's pretty much what I needed. My test case would be for the former case there (actually hanging). I was more looking for a specific case that shouldn't hang. How long is the timeout? > > Ideally you're spending your time on changes that visibly improve one of > the major benchmarks we're tracking anyway. In that case, there's no need > to add a separate performance test; but please do add correctness tests for > the feature you're adding/modifying. > > Other than that we usually don't have fine-grained performance tests. > Optimizing for micro-benchmarks is easy but often pointless. Also, it's > important to realize that the test suite is supposed to assert correctness > while finishing as quickly as possible, whereas measuring performance > typically requires running the tests in question for a considerable amount > of time to get reliable results. > > > On Thu, Sep 25, 2014 at 4:19 PM, Isiah Meadows <impi...@gmail.com > <javascript:>> wrote: > >> How's the best way to test for a performance regression (as in "this test >> should not hang")? >> On Sep 23, 2014 1:28 AM, "Ben Noordhuis" <in...@bnoordhuis.nl >> <javascript:>> wrote: >> >>> On Tue, Sep 23, 2014 at 1:48 AM, Isiah Meadows <impi...@gmail.com >>> <javascript:>> wrote: >>> > What I need to do for a specific unit test is to run a specific method >>> + >>> > arguments, and if it takes too long, stop the call mid-cycle and fail >>> the >>> > test. This is for a performance-related unit test for my patch (which >>> is a >>> > perf patch itself, anyways). Is this possible, and if so, how would I >>> do it? >>> > I'm not nearly as well versed in C++ as I am in JavaScript, but I could >>> > write it in C++ if I had to. >>> >>> It sounds like you would need to call V8::TerminateExecution() from a >>> watchdog thread. That said, tests that rely on wall clock time are >>> unreliable and will almost certainly get rejected during code review. >>> >>> -- >>> >> -- -- v8-users mailing list v8-users@googlegroups.com http://groups.google.com/group/v8-users --- You received this message because you are subscribed to the Google Groups "v8-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to v8-users+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.