On Wed, Apr 18, 2012 at 19:15, Blair Zajac <bl...@orcaware.com> wrote: > On 04/18/2012 03:45 PM, Greg Stein wrote: >... >> Yup, I understood. Sorry that I wasn't clear: I meant in our >> serialization code, shouldn't we use the "proper" functions rather >> than raw sha1 digests? Is there any way to switch to them at a >> higher/semantic level? >> >> I haven't looked at that stuff, but I'm going to guess repositories >> now exist with raw sha1 digests. Is there a format type in there? Can >> we start writing "csum" and svn_checksum_serialize() into the skel? >> And then read raw md5, raw sha1, or a serialized checksum? > > I'm not familiar with this part of the code either (not having looked at the > svn_skel.h before) but that make sense. The code could look for a raw > digest or a $name$ and then use that. > > It doesn't look like it would be too hard. Because the skel has a len, if > it's equal to APR_SHA1_DIGESTSIZE then you would use the raw digest, > otherwise use svn_checksum_deserialize(). > > Sounds like a 1.8 repository upgrade though, since once you wrote a $sha1$ > style string, older Subversion's wouldn't parse it directly.
Oh. Good point! > I won't be taking this on, I've still got a memory lifetime issue I'm > debugging and then a svn commit thread pool to write to support 4 > commits/sec from remote clients that are consuming all threads from our RPC > thread pool and killing readers. hehe... that'll take you what... an hour? ;-) Cheers, -g