[RFC] Knit format 2
John Arbash Meinel
john at arbash-meinel.com
Sun Sep 10 04:31:43 BST 2006
-----BEGIN PGP SIGNED MESSAGE-----
Aaron Bentley wrote:
> John Arbash Meinel wrote:
>>> Aaron Bentley wrote:
>>> I think we just need to factor this sort of thing into a
>>> 'get_revision_trees' helper, which will do that sort of thing, and
>>> return sort of a 'best case' read size. I think you can stick with 500
>>> for now, because we aren't dealing with the freebsd ports tree (yet).
> I don't recall a significant difference between 100 and 500 here, so
> I've dropped it to 100.
>>> But if you are going to do any work, then having one that inspects the
>>> inventory size would be my preference.
> Well, I don't want to get into too many side issues. Remember this is
> already a side issue to nested-trees!
>>> Because you have the same if revision_id is None: revision_id = ...
>>> check in iter_rev_trees(). (You aren't reading the revision trees directly)
> Doh! Sorry I didn't read your earlier comment carefully enough.
>>>> I see two options:
>>>> 1. Depend on the inventory format, not the repository format
>>> I would be okay with you doing it in lots of different ways. But it
>>> should probably still be a white-list style not blacklist. Because
>>> fileids_affected... is really a violation of api constraints.
> Yes. It's a hack I'm not fond of. Anyhow, I've made it an attribute of
>>> We can't really change existing formats, because they are a disk
>>> representation, and older bzr clients would puke.
>>> We could upgrade our internals, and then in the next format start using
>>> trailing newlines. But the only way to tell that you can do that, is by
>>> having a flag in the format, and you might as well just have that flag
>>> be a '\n' at the end of the line. :)
> Well, I'll take the easy way out for now. But I can easily imagine this
> class of mistake being repeated.
> I've attached my changes. If you prefer the complete diff, I can
> provide that, too.
I'm happy enough with this. It isn't the default format, and it is a
necessary stepping stone to get the nested trees stuff working properly.
So +1 from me.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
-----END PGP SIGNATURE-----
More information about the bazaar