[PACKS] Performance opportunities.

[Robert Collins]

On Thu, 2007-08-30 at 19:58 -0400, Aaron Bentley wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Robert Collins wrote:
> > Well its still layered on top of _iter_changes; but it does have the
> > three-valued approach that John speculates about. I agree that
> > _iter_changes need to change to reflect this too.
> 
> The only case where it is expensive to determine whether a file has
> changed is when the stat value has changed, but the st_size has not.
> Are you claiming that is a significant case?  Do you have data?
> 
> It's frustrating to see you guys talking about this over and over again,
> when I specifically designed this API to be cheap in the vast majority
> of cases.  No one has yet claimed that my evaluation was wrong.
> Instead, you mostly claim that it always requires SHA1s, which is
> patently not true.

Well let me refine the claim then; dirstate's implementation probably is
overaggressive on sha1 usage and 'simply' a bug - but one exacerbated by
the current layer interactions. Secondly there is a corner case where
the sha1 will be calculated redundantly and that can be expensive. The
case where I would expect that to happen is for things like .isos and
other binary/database style files which often work in data rounded to a
block/cluster.

-Rob


-- 
GPG key available at: <http://www.robertcollins.net/keys.txt>.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20070831/dcda3491/attachment-0001.pgp