On Fri, May 1, 2015 at 12:39 AM, Jelmer Vernooij <[email protected]> wrote:
> Actually, I mostly meant to say that treating them as context managers
> is one approach, but there are other options. This is the kind of
> thing we can't easily change once it's been introduced as it it's a
> significant chunk of the public API, so we should be careful.
>
Ah. So maybe we just add Repo.close (no new context managers in the
library) and then we use contextlib.closing in the tests to do
with closing(repo):
....
What kind of operations are we trying to optimize exactly? Which
> Dulwich users (with what access patterns?) are we trying to help?
>
> I'm not convinced statting for existing (but missing) loose objects is
> actually a significant problem at the moment. In what cases would you
> stat for the same object SHA multiple times *and* want avoid a stat?
>
> Not reading/statting the packlist multiple times seems to me like it
> would definitely be a win. Do we need API changes for that though, or
> could we e.g. just use inotify to watch it?
>
I was not aware of the statting for loose objects. I was just thinking
about the statting for pack list.
I'm not sure about this. Caching adds complications; let's start a
> separate thread about what our goals are with this, and the pros/cons
> of the various options.
>
I would personally be happy to leave the caching as is, and only look at
ensuring that we close all files.
Regards,
Gary
_______________________________________________
Mailing list: https://launchpad.net/~dulwich-users
Post to : [email protected]
Unsubscribe : https://launchpad.net/~dulwich-users
More help : https://help.launchpad.net/ListHelp
