> Indeed, mode-line updates are *very* frequent (more than once per
> command), so we really want to avoid running processes at that point.

I though the new implementation of the vc-git mode line doesn't use
vc-working-revision, so in this specific case, making vc-working-revision
slower won't hurt us there.

> I wonder if that's true. Caching in a property is obviously fast, and
process calls are
necessarily an order of magnitude slower (and slower still on certain
platforms).

True, but you have to check that the cache is invalid somehow. If you want
to be super correct, a rev-parse is probably the fastest way. I suppose it
would be pretty fast to check the mtime of the .git directory inside emacs
and invalidate that when it changes. Although filesystem timestamps usually
make me nervous, it's probably okay in this instance?



On Thu, Sep 3, 2015 at 9:17 AM, Stefan Monnier <monnier@iro.umontreal.ca>
wrote:

> > Maybe you should write a patch and test it on Windows (or have someone
> else
> > do it), and see whether the mode-line updates don't become
> > perceptibly slower.
>
> Indeed, mode-line updates are *very* frequent (more than once per
> command), so we really want to avoid running processes at that point.
>
>
>         Stefan
>