On Fri, Mar 13, 2015 at 01:50:33PM +1300, Robert O'Callahan wrote: > On Fri, Mar 13, 2015 at 12:34 PM, Seth Fowler <[email protected]> wrote: > > > I guess (but don’t know for sure) that recording RR data for every test > > that gets run might be too expensive. > > > It probably wouldn't be too expensive. The runtime overhead is low; the > main cost is trace storage, but we can delete traces that don't reproduce > bugs. > > AFAIK the limiting factors for running rr in our test infrastructure are: > 1) rr only supports x86(64) Linux, so for bugs on other platforms you'd be > out of luck. Hopefully we'll have ARM support at some point, but non-Linux > isn't going to happen. > 2) Our Linux tests run on EC2 and Amazon doesn't enable perfcounter > virtualization for EC2. This could change anytime (it's a configuration > change), but probably won't. > 3) Some bugs might not reproduce when run under rr. > 4) We'd need to figure out how make rr results available to developers. It > would probably work to move the traces to an identically-configured VM > running on the same CPU microarchitecture and run rr replay there. > So I don't think we're going to have rr in our real test infrastructure > anytime soon.
5) Not all EC2 hosts have the right CPU microarchitecture. FWIW. Mike _______________________________________________ dev-platform mailing list [email protected] https://lists.mozilla.org/listinfo/dev-platform
