Re: [PATCH v10 3/4] add `config_set` API for caching config-like files
I think my proposal is a fix, and yours is a workaround. Semantically, calling setup_git_directory changes the config, so it should invalidate the cache. Your proposal consists in not filling-in the cache before it is invalidated, which works, but isn't very future-proof: if anybody fills-in the cache before the setup in the future, it will break it.
Now, both options can go together, then moving the check_pager_config call after setup_git_directory would avoid re-filling-in the cache. Intuitively, I'd say it's even needed in case git is started from a subdirectory and the .git/config contains pager-relevant stuff, but there might be complications.
----- Original Message -----
Show 43 quoted lines
>
>
> On 7/18/2014 3:10 PM, Matthieu Moy wrote:
> > ----- Original Message -----
> >> Documentation/technical/api-config.txt | 137 +++++++++++++++++
> >> cache.h | 30 ++++
> >> config.c | 263
> >> +++++++++++++++++++++++++++++++++
> >> 3 files changed, 430 insertions(+)
> >
> > I think the added call to git_config_clear() I proposed yesterday in
> > setup_git_directory_gently_1 should be part of this patch (with the
> > associated comment), just like this call:
> >
>
> Oh, my bad, I thought you meant that I incorporate it with the
> git_config_raw() patch.
> I wanted to ask, can we call setup_git_directory_gently() earlier in
> execv_dashed_external()
> in git.c, which calls check_pager_config() for the first time which causes
> the
> incomplete cache to formed.
> If we can do it, we won't have to clear the cache every time
> setup_git_directory_gently_1()
> is called.
>
> >> @@ -1707,6 +1967,9 @@ int git_config_set_multivar_in_file(const char
> >> *config_filename,
> >> lock = NULL;
> >> ret = 0;
> >>
> >> + /* Invalidate the config cache */
> >> + git_config_clear();
> >> +
> >> out_free:
> >> if (lock)
> >> rollback_lock_file(lock);
> >
> > I have limited access to my email and no way to apply the patches today, so
> > I can't do a detailed review. But other than the remark above, I guess the
> > patch series is now all right and ready to cook in pu.
> >
>
--
Matthieu Moy
http://www-verimag.imag.fr/~moy/
Matthieu Moy <matthieu.moy@grenoble-inp.fr>