[PATCH 1 of 3] dirstate: don't check state of subrepo directories

[Augie Fackler]

On Dec 31, 2009, at 4:15 PM, Matt Mackall wrote:

> On Thu, 2009-12-31 at 14:36 -0600, durin42 at gmail.com wrote:
>> # HG changeset patch
>> # User Augie Fackler <durin42 at gmail.com>
>> # Date 1261857501 21600
>> # Node ID 68d80c9814d720f2b7d8b2f39008ae6547a5b128
>> # Parent  513c89a60f203a8387d8dfa6f2dc67b21725790a
>> dirstate: don't check state of subrepo directories
>
>> -    def status(self, match, ignored, clean, unknown):
>> +    def status(self, match, ignored, clean, unknown, skip=[]):
>
> When do we ever not want to skip subrepos? And why not call this arg
> subrepos so we don't abuse it for something else?

Fair enough.

>> -            s = self.dirstate.status(match, listignored,  
>> listclean, listunknown)
>> +            s = self.dirstate.status(match, listignored,  
>> listclean, listunknown,
>> +                                     skip=set(ctx1.substate.keys() +
>> +                                              ctx2.substate.keys()))
>
> Why are we passing both contexts' substate info down?

Because dirstate is kind of hairy, and I probably got confused at some  
point.

> The only one that is relevant to status in the working directory is  
> the workingctx, yes?

Yes, agreed.

> And I think that's always ctx2:
>
>        working = ctx2.rev() is None

er, wha? I'm pretty sure it's ctx1. The logic in here is twisty.

> Backing up a bit, it'd be nice to wire this up when we create the
> dirstate object:
>
>    @propertycache
>    def dirstate(self):
>        return dirstate.dirstate(self.opener, self.ui, self.root,
> skiplist)
>
> Then status/walk callers never have to think/know about subrepos,  
> which
> is nice. Maybe there aren't enough callers to matter?

This change actually touches all the callers, I grepped around.

> But this implies reading .hgsub (possibly unnecessarily) at create  
> time.

Yeah, localrepo has already parsed .hgsub.

> Which we don't really want to do. Nor do we want dirstate to hold a
> reference to localrepo, so passing in a callback tricky.
>
> Alternately, we could teach dirstate internally about parsing .hgsub
> files similar to how it already knows about .hgignore files.

I suppose that would work, and then it'd turn into the same skip list  
as now? And then we'll still disable inotify when it's non-empty until  
inotify learns?

If it seems like the cleanest way, I'll look at having dirstate  
parse .hgsub directly without a context. That seems like the most  
viable alternative (right now) to how it's implemented in this change.

> -- 
> http://selenic.com : development and support for Mercurial and Linux
>
>