Comment 23 for bug 415508

2009/8/24 Robert Collins <email address hidden>:
> So, its worth noting I think, that we got significant performance
> increases by path_content_summary, because it avoided path computation
> multiple times inside tree; the new commit code path largely addresses
> that :- we're actually rolling back a change here ;)

Right, I realized that after posting last night. We need to make sure
that if the canonical hash is known, we use it.

The problems in path_content_summary can be seen as:

 * it does not consistently use the canonical form
 * it's contract is meant to be "return the size, and return the SHA1
if you can do that cheaply", but in fact it needs to be allowed to not
return the size if it can't do it cheaply

I'm adding some developer documentation about content filtering now to
make some of the issues discussed here more permanent and explicit.

--
Martin <http://launchpad.net/~mbp/>