On Tue, May 07, 2019 at 12:54:28PM +0200, Ævar Arnfjörð Bjarmason wrote: > So here as a 6-parter given some of the feedback on v2. Maybe Jeff > still hates it :), but this time around some of the changes are split > up and should be easier to understand in isolation, as well as some > more "noticed while I was at it" things fixed. Thanks, the split did make things easier to understand. These all look reasonable to me, and I confirmed by re-running a few of the problematic perf tests that everything is behaving as expected after the patches. > This series ends with outright forbidding the user from directly > setting GIT_TEST_INSTALLED. As discussed it makes things easier for > us, and as noted in 1/6 the demand for that in the wild seems > non-existent, since the way we've been documenting how you could do > that with an environment variable has been broken since 2012. I think this is reasonable, especially with the feature in patch 6 to tell the caller they're doing it wrong, instead of just silently producing nonsense results. Junio, if you haven't been following closely, this can replace my patch from jk/perf-installed-fix. -Peff
