The default method does much more than just a function call. It has
to create the subrequest, which does some data conversion, go through
the apache handler chain, then undo the data conversion. The same
step is replaced with a lookup (should be very fast) and a function
call.
I'm planning on testing something very similar to what you suggest.
On 4/12/07, David James <james@cs.toronto.edu> wrote:
> Hi Josh,
>
> With your patch, a request for a log message would still generate 1
> lookup per file affected in the revision. If a revision affects, say,
> a hundred thousand files, your patch would still trigger a hundred
> thousand provider lookups and a hundred thousand calls to said
> provider when a user requests to look at the log message.
>
> What is the overhead of a provider lookup and call as compared to a
> regular function call? If it's not a lot more expensive then maybe
> this is not an issue.
>
> Try importing a hundred thousand files in a single revision into a
> repository, and then running 'svn log' on any file on the repository.
> If you do this using ra_svn you'll find that there are no performance
> issues, even with authz enabled. Without your patch, ra_dav takes
> about 30-90 minutes to calculate all of the authz lookups in this
> scenario.
>
> I'd be curious to see how much your patch helps.
>
> Cheers,
>
> David
>
--
Josh Gilkerson
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Apr 12 21:12:27 2007