Ryan Schmidt wrote on Fri, Oct 29, 2010 at 18:26:25 -0500: > On Oct 29, 2010, at 17:11, Daniel Shahaf wrote: > > > Short version: it's a bug in Google's backend, and I can reproduce it > > with neon but not with serf. > > > % $svn log -v -r2967 http://pure-lang.googlecode.com/svn/trunk > > --config-option servers:global:http-library=serf > > ------------------------------------------------------------------------ > > % echo $? > > 0 > > % $svn log -v -r2967 http://pure-lang.googlecode.com/svn/trunk > > --config-option servers:global:http-library=neon > > ... > > subversion/libsvn_ra_neon/util.c:773: (apr_err=160004) > > svn: Observed changes didn't match count > > % echo %? > > 1 > > % ./tools/dev/which-error.py 160004 > > 00160004 SVN_ERR_FS_CORRUPT > > % $svn --version | head -n2 > > svn, version 1.7.0 (dev build) > > compiled Oct 27 2010, 15:38:01 > > % > > I agree using serf instead of neon makes the error message go away. > But it also doesn't actually show the list of files affected by the > commit like it should. See "svn diff --summarize -r2967" for all the > files that should be shown there. >
More specifically, it behaves as if there are no relevant revisions to show: [[[ % $svn log -vr1000000 http://svn.eu.apache.org/repos/asf/subversion ------------------------------------------------------------------------ % $svn log -vr1000000 http://svn.eu.apache.org/repos/asf ------------------------------------------------------------------------ r1000000 | yonik | 2010-09-22 16:48:10 +0200 (Wed, 22 Sep 2010) | 1 line Changed paths: M /lucene/dev/trunk/solr/CHANGES.txt M /lucene/dev/trunk/solr/src/java/org/apache/solr/search/FunctionQParser.java M /lucene/dev/trunk/solr/src/java/org/apache/solr/search/QueryParsing.java M /lucene/dev/trunk/solr/src/test/org/apache/solr/search/FunctionQParserTest.java M /lucene/dev/trunk/solr/src/test/org/apache/solr/search/function/TestFunctionQuery.java SOLR-2128: full param substitution for function queries ------------------------------------------------------------------------ % ]]] There shouldn't be any authz restrictions on this repository, right? > Also, even when I change http-library to serf in the servers config > file, svnsync still shows the "Observed changes didn't match count" > error. Does svnsync not use this library setting? > Examining the code shows that the apr_hash_t *config is passed down to svn_ra_open4(), and upon trying I cannot reproduce this: [[[ % $svnsync init http://localhost:8081/t/{r2,r1} DBG: ra_loader.c: 493: vtable='Module for accessing a repository via WebDAV protocol using serf.' url='http://localhost:8081/t/r2' DBG: ra_loader.c: 493: vtable='Module for accessing a repository via WebDAV protocol using serf.' url='http://localhost:8081/t/r1' Copied properties for revision 0. % mkdir cfg % cat > cfg/servers [global] http-library = neon % $svnsync sync http://localhost:8081/t/r2 --config-dir cfg DBG: ra_loader.c: 493: vtable='Module for accessing a repository via WebDAV protocol using Neon.' url='http://localhost:8081/t/r2' DBG: ra_loader.c: 493: vtable='Module for accessing a repository via WebDAV protocol using Neon.' url='http://localhost:8081/t/r1' Transmitting file data ............ Committed revision 1. Copied properties for revision 1. ]]] > > >> I'm unsure what we should try from here. Possibly this error is unique > >> to Google Code's Subversion backend, > > > > The error code confirms your suspicion. Furthermore, the word > > "Observed" doesn't appear in Subversion's source code. > > I suppose we should report the problem to Google Code and hope they > can uncorrupt the repository somehow? Yes. SVN_ERR_FS_CORRUPT justifies reporting this to the Google admins. > Apparently a couple dozen > subsequent revisions see the same error, through r2993. >
