On Dec 4, 2004, at 11:33 AM, Philip Martin wrote:
>
> So that about 50% slower but I think the correctness arguments justify
> that regression.
Ah, that's about what I expected. The whole reason for the cache was
to decrease the number of turnarounds when committing a single file.
ra_dav needs to issue a CHECKOUT of a version-resource before doing a
PUT of it, and if no version-resource cache is available, then a few
PROPFINDS need to happen to convert (rev, path) into a version resource
url.
It's funny, you know. There's quite a lot of benefit to making
mod_dav_svn interoperable with generic DAV or deltaV clients... like
the current autoversioning feature. But AFAIK, there's not a *single*
deltaV server out there in the world, so the goal of making the svn
client "a well behaved deltaV client" seems pretty useless these days.
Do we ever expect svn clients to operate against some future deltaV
server?
If we decided to drop that goal, the svn client would be allowed to
"know" the format of mod_dav_svn's urls...
- it could convert (rev, path) into version-resource-url with no
turnarounds
- no more need for a cache
- a whole bunch of ra_dav operations would lose many network
turnarounds.
>
> Perhaps it would be possible to detect a 1.2 server and have the
> client conditionally request the resource walk?
>
Hm, not sure how to do this. I suppose a 1.2 client could send a
special If: header in the REPORT request, and only a 1.2 server would
understand it to mean, "ah, I guess I'll send the resource walk."
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Dec 4 19:29:07 2004