The "svn status" result is cached by _subversion in $_cache_svn_status, but the cache is not updated after a modification of the files (which may lead to a change in the "svn status" result). Therefore later svn completions sometimes fail. IMHO, it would be sufficient to clear the cache at the next prompt in practice. The files may be modified by another process while a svn command is still being typed, but this wouldn't be common.
Maybe use the caching layer and expire it after 10 seconds?
[moved to -workers] Repeating what I said at http://bugs.debian.org/342347 , maybe a very short cache expiration is the way to go.
[moved to -workers] Repeating what I said at http://bugs.debian.org/342347 , maybe a very short cache expiration is the way to go.
Note that this is not sufficient if you use another shell session to call svn (I frequently do that). Yes, maybe. Or can't something like FAM[*] optionally be used when supported? (Note: if it is used, it should be a run-time option, as I'm not sure that FAM is OK with all NFS servers.) [*] http://en.wikipedia.org/wiki/File_alteration_monitor