On Wed, 21 Sep 2016 13:26:31 -0700, Durham Goode wrote: d> On 9/21/16 1:07 PM, Arun Kulshreshtha wrote: > > On 9/21/16, 7:34 AM, "Yuya Nishihara" <y...@tcha.org> wrote: > > > Additionally, we’d need to wrap other callsites of > > > _runcommand(), such as _checkshellalias(), to maintain the existing > > behavior. > > > > Good point. Given that we want to start profiling as early as > > possible, I think > > it's better to test profiling.enabled before extensions.loadall(lui), > > and test > > it again after parsing command options. > > > > CC’ing Durham on this. I considered doing this, but during our (offline) > > discussion about this, it seemed > > like it was undesirable to profile uisetup() for each extension, which is > > why I placed maybeprofile() after > > extensions.loadall(). However, if we do indeed want to start profiling as > > early as possible, then I’ll try enabling > > profiling as early as possible in _dispatch() and remove the other > > callsites. > The only reason I hesitated to wrap uisetup and extsetup is that > extensions may modify the profile logic (like we have a statprofext > extension that configures statprof), so wrapping early may make that > impossible.
Yeah, that's a valid point. It makes some sense to start profiling after ui/extseutp(). Maybe --debugger has the same chicken-and-egg issue. _______________________________________________ Mercurial-devel mailing list Mercurial-devel@mercurial-scm.org https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel