[PATCH] histedit: add support for exec/x command to histedit (issue4036)

[Yuya Nishihara]

On Sun, 25 Oct 2015 16:33:15 +0000, Piotr Listkiewicz wrote:
> >
> >  My point here is that if some command changed stuff (new commit,
> >
> >     bookmark location etc), some of it might not be detected by the
> >     current process that already parsed them from disk and naively
> >     assume they did not changed since it hold the lock.
> >     I would like to see some investigation to know if we need to do
> >     something special (or if it is already handled, for example by the
> >     lock). to make sure we are safe here.
> >
> > We should talk about this, i still have no idea how can i check this or
> > what to do with it
> > Sid, Foozy and yuya are good people discuss this
> >
> From my investigation and talk with yuya i think that all caches are
> cleared:
> 
> finally:
>     self.state.wlock = repo.wlock(times=wlockheld)
>     self.state.lock   = repo.lock(times=lockheld)
> 
> repo.wlock invoke lock constructor with acquirefn callback set to
> localrepo.invalidatedirstate.
> 
> repo.lock invoke lock constructor with acquirefn callback set to
> localrepo.invalidate.
> 
> invalidateall function in localrepository has invocation only to invalidate
> and invalidatedirstate.
> 
> So if acquiring those two locks dont clear all caches that would mean that
> localrepository.invalidateall has bug right now.

One thing I forgot to mention, invalidateall() is wrapped by MQ. It is
necessary for command server to clear mq status and mq path. So "histedit exec"
might need to call invalidateall().

Regards,