Re: "SVNAuthorizationShortCircuit or something similar"

There are some very obvious bugs in the previous patch, but with
fixes, it seems to work. I still want to do more correctness testing
and post the patch this weekend. Performance seems to be vastly
improved.

Getting the log for a commit with ~25000 additions:
"stock" ubuntu module: 12.6 s
patched module, default: 12.6 s
patched module, bypass: 0.9 s

So it looks like a speed up of >10x on this size operation.

Regarding caching the provider lookup, I want to run this by you guys
and make sure I'm not doing something I shouldn't be. I added a
static variable in mod_dav_svn.c that holds the pointer. In the
configuration callback, if bypass is configured and it is still NULL,
I do the lookup and set it. Seems to work, but I'm not an apache
guru.

Another point I am not completely sure about, using the subrequest
builds a uri with escaping, etc. The authz handler then splits the
uri, but it doesn't appear that it does any unescaping. Does the
subrequest mechanism do this or does the lower authz function expect
an uri-escaped path? And a related note, it seems that the revision
is ignored by the authz mechanism. This makes sense given the authz
file format, but again, I want to make sure.

-- 
Josh Gilkerson
Software Engineer
Google, Inc · MV-1600 Plymouth (HQ)
+1 (650) 253-1667 direct
+1 (859) 608-7827 cell
jwg@google.com
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org