Is this a fix or a work-around? i.e., if you set MAX_OPEN_HISTORIES to,
say, 2, can you then reproduce the failure you saw before?
It's just not clear to me whether MAX_OPEN_HISTORIES is in fact an
assumption that "no file will be 'svn mv'd more than 32 times" or
similar.
Kevin Radke wrote on Wed, Jan 26, 2011 at 16:59:54 -0600:
> This also needs a back-port to the 1.6.x branch.
>
> [[[
> Fix issue #3789: Correctly ignore missing locations when a renamed file
> has more than MAX_OPEN_HISTORIES.
>
> * subversion/libsvn_repos/log.c
> (get_path_histories): Ignore more bogus repository locations to restore
> pre-1.6.15 log -g behavior. This fixes client
> chunk
> errors introduced by r1028108.
> ]]]
>
>
> Index: subversion/libsvn_repos/log.c
> ===================================================================
> --- subversion/libsvn_repos/log.c (revision 1063904)
> +++ subversion/libsvn_repos/log.c (working copy)
> @@ -1052,6 +1052,7 @@
> {
> svn_fs_root_t *root;
> apr_pool_t *iterpool;
> + svn_error_t *err;
> int i;
>
> /* Create a history object for each path so we can walk through
> @@ -1093,7 +1094,6 @@
>
> if (i < MAX_OPEN_HISTORIES)
> {
> - svn_error_t *err;
> err = svn_fs_node_history(&info->hist, root, this_path, pool);
> if (err
> && ignore_missing_locations
> @@ -1115,10 +1115,20 @@
> info->newpool = NULL;
> }
>
> - SVN_ERR(get_history(info, fs,
> - strict_node_history,
> - authz_read_func, authz_read_baton,
> - hist_start, pool));
> + err = get_history(info, fs,
> + strict_node_history,
> + authz_read_func, authz_read_baton,
> + hist_start, pool);
> + if (err
> + && ignore_missing_locations
> + && (err->apr_err == SVN_ERR_FS_NOT_FOUND ||
> + err->apr_err == SVN_ERR_FS_NOT_DIRECTORY ||
> + err->apr_err == SVN_ERR_FS_NO_SUCH_REVISION))
> + {
> + svn_error_clear(err);
> + continue;
> + }
> + SVN_ERR(err);
> APR_ARRAY_PUSH(*histories, struct path_info *) = info;
> }
> svn_pool_destroy(iterpool);
