So I've started the merge of svn_kind_t into svn_node_kind_t and have
run across a more fundamental kind of conflict that I can't resolve
without the help of the authors of the tree-conflict storage.
Here's the problem: the WC code has a mechanism for converting
enumeration values used in code to tokens used by the database. This
mechanism uses a conversion table. For the vast majority of uses, the
table looks like this:
[from subversion/libsvn_wc/token-map.h]
static const svn_token_map_t kind_map[] = {
{ "file", svn_kind_file }, /* MAP_FILE */
{ "dir", svn_kind_dir }, /* MAP_DIR */
{ "symlink", svn_kind_symlink }, /* MAP_SYMLINK */
{ "unknown", svn_kind_unknown }, /* MAP_UNKNOWN */
{ NULL }
};
However, for tree conflict storage, we have:
[from subversion/libsvn_wc/tree_conflicts.c]
static const svn_token_map_t node_kind_map[] =
{
{ "none", svn_node_none },
{ "file", svn_node_file },
{ "dir", svn_node_dir },
{ "", svn_node_unknown },
{ NULL }
};
Note that the first table is for svn_kind_t, whereas the second is for
svn_node_kind_t (which supposedly shouldn't have been used in the WC
code in the first place); but that's not the issue, the biggest problem
is the difference in mapping svn_kind_unknown and svn_node_unknown.
I could leave the difference as-is, but frankly it seems silly to be
inconsistent in the way we represent essentially the same things in the
database. In my opinion, this calls for a WC-version bump before we
branch 1.8.
Other opinions please?
-- Brane
--
Branko Čibej
Director of Subversion | WANdisco | www.wandisco.com
