On Tue, Jan 25, 2011 at 10:34:43AM -0600, Kevin Radke wrote:
> (I'm moving this conversation from users to dev, since I have
> convinced myself a regression was introduced in r1028108)
> See: http://svn.haxx.se/users/archive-2010-12/0265.shtml
>
> log -v -g --xml http://server/repo/path commands will now fail for
> "complex" histories that contain file renames.
>
> The client sees:
> svn: REPORT of '/repo/!svn/bc/1234/path/in/repo': Could not read chunk
> size: connection was closed by server (http://server)
>
> The server logs:
> [Wed Dec 15 15:48:18 2010] [error] [client 192.168.1.1] File not
> found: revision 5678, path '/path/in/repo/file.txt' [404, #160013]
>
> (The file was named oldfilename.txt in r5678, because it was renamed
> in r7890. Something isn't using the correct name when more than
> MAX_OPEN_HISTORIES have been found.)
>
> The only source file modified in this commit was subversion/libsvn_reops/log.c
>
> A few questions:
>
> 1) Does setting info->oldpool and info->newpool to NULL around lines
> 1113 potentially leak memory?
> 2) What is info->first_time used for? It seems to always be set to
> true in the loop in get_path_histories() and then reset in
> get_history()
> 3) Increasing MAX_OPEN_HISTORIES to 128 "fixes" my test repository,
> but isn't the true fix.
Thanks for bringing this to dev@.
If it's not too much of a bother, could you also file an issue for this
and set the target milestone to 1.7.0 so we don't forget about it?
This milestone doesn't mean that a fix for 1.6.x won't be made.
It just exploits the fact that currently most people are looking at
issues scheduled for 1.7.0 :)
Thanks,
Stefan
Received on 2011-01-25 20:14:42 CET