On 1/4/2018 6:06 PM, Jeff King wrote:
On Wed, Jan 03, 2018 at 09:47:28PM +0000, Jeff Hostetler wrote:
Config values of true and false control non-porcelain formats
for compatibility reasons as previously discussed. In the
last commit I added a new value of 2 for the config setting
to allow porcelain formats to inherit the new setting. I've
marked this experimental for now or so that we can discuss
it.
I'm mildly negative on this "level 2" config. If influencing the
porcelain via config creates compatibility headaches, then why would we
allow it here? And if it doesn't, then why do we need to protect against
it? This seems to exist in a funny middle ground that cannot decide
whether it is bad or not.
It's like we're inserting a foot-gun, but putting it just far enough out
of reach that we can blame the user when they shoot themselves with it.
Is there a compelling use case for this? From the previous discussion,
this is the strawman I came up with:
Scripted callers like Visual Studio don't want to unconditionally pass
--no-ahead-behind, because it makes sense only for large repositories
(and small ones would prefer the more exact answer, if we can get it
cheaply). So we'd like the user to trigger "this is large" on a
per-repo basis, and accept the consequences of possibly broken
porcelain callers.
I think we could have the best of both worlds, though, if the existing
config option were coupled with a command-line option to say "yes, I
understand no-ahead-behind, so use it for the porcelain if applicable".
IOW, the user does:
git config status.aheadbehind false
and VS does:
git status --ahead-behind=maybe
and together both sides have assented to the "quick" thing.
-Peff
I kinda trying to serve 2 masters here. As we discussed earlier, we
don't want config options to change porcelain formats, hence the
true/false thing only affecting non-porcelain formats. On the other
hand, VS and git.exe are on different release schedules. Normally,
I'd just have VS emit a "git status --no-ahead-behind --porcelain=v2"
and be done, but that requires that git.exe gets updated before VS.
We do control some of that, but if VS gets updated first, that causes
an error, whereas "git -c status.aheadbehind=<x> status --porcelain=v2"
does not. It is respected if/when git is updated and ignored until
then. Likewise, if they update git first, we can tell them to set a
config setting on the repo and inherit it for porcelain v2 output
without VS knowing about it. Sorry, if that's too much detail.
It is OK with me if we omit the last commit in the patch series (that
does the experimental =2 extension) and I'll deal with this separately
(maybe differently) in the gvfs fork.
Thanks,
Jeff
