2012/3/22 Julian Foad <julianf...@btopenworld.com>: > Sérgio Basto wrote: > >> Philip Martin wrote: >>> Or you could rebuild the current APR with a >>> one-line patch to set ht->seed to 0 in apr_hash.c:apr_hash_make. >> >> I think when subversion receive the result of this apr_function , we >> should do the sort , ie , subversion should not depend on sort or not >> sort of the apr . >> >> So where in source code we have the return of apr function ? > > There is not just one place, there are many places, even within the diff > code. Start in subversion/svn/diff-cmd.c, at svn_cl__diff(), and follow the > function calls down. For a WC-to-WC diff, these are the main calls: > > svn_cl__diff > | > svn_client_diff[_summarize]6, svn_client_diff[_summarize]_peg2 > | > svn_wc_diff6 > | > svn_wc__internal_walk_status > | > get_dir_status > | > /* Walk all the children of this directory. */ > for (hi = apr_hash_first(subpool, all_children); > hi; hi = apr_hash_next(hi)) > > This place determines the order of entries within a directory, for a WC-to-WC > diff. That's just one of the code paths involved in diff. There are several > others, depending on the kind of diff. For diffs against the repository, it > is more complex. Also > look for code that determines the order of property changes.
This reminds me of something: it may be coincidence, but up until now (until 1.7.x that is) 'svnlook changed -rX' and 'svnlook diff -rX' presented the files in the same order. I like this behavior, because it's nice when composing post-commit emails: the list of changed files on top, followed by the actual diff ... both following the same order. Anyway, no biggie, all I'm saying is: +1 for introducing some stable ordering here ... -- Johan