Re: [BUG] git status doesn't handle submodules properly on OSX
- From
Jeff King <peff@peff.net>
- Date
- Oct 16, 2008, 14:18 UTC
- Message-ID
- <20081016141812.GA30026@sigill.intra.peff.net>
- In-Reply-To
- <8199b7ae4c441c4311045141ddaaa36f.squirrel@webmail.highteq.net>
On Thu, Oct 16, 2008 at 02:30:00PM +0200, Lars Hoss wrote:
Show 6 quoted lines
> The commit was on Jun 5, 14:47:50 by Marius Storm-Olsen and the relevant > file is wt-status.c. > > Ok, I think I found the issue. I enabled showUntrackedFiles in my gitconfig: > > status.showUntrackedFiles = all
Ah, OK. I see what is going on. All code paths call the read_directory infrastructure to find untracked files. If status.showUntrackedFiles is "normal", then we set dir.show_other_directories, to indicate that we want to see the directories, but not their constituent files.
If status.showuntrackedfiles is set to "all", then we don't set the show_other_directories flag, because we want each file. But the code in dir.c:treat_directory uses the "show_other_directories" flag to say "oh, we're just interested in untracked files" and decide whether to ignore gitlinks.
Meaning that we must still filter the results of read_directory based on the cache. And indeed, this is what "git ls-files -o" does, as explained in 5698454e (Fix some "git ls-files -o" fallout from gitlinks). It's also what the code in wt_status_print_untracked is _supposed_ to do, but it was never updated to handle this case when git-ls-files was. Which is probably my fault a long time ago for cutting and pasting the 5 lines of "is this thing in the cache" when writing wt-status.c.
So the quick fix is to re-cut-and-paste the code:
diff --git a/wt-status.c b/wt-status.c index d2eac36..792d5f1 100644 --- a/wt-status.c +++ b/wt-status.c @@ -280,10 +280,14 @@ static void wt_status_print_untracked(struct wt_status *s) /* check for matching entry, which is unmerged; lifted from * builtin-ls-files:show_other_files */ struct dir_entry *ent = dir.entries[i]; - int pos = cache_name_pos(ent->name, ent->len); + int len, pos; + len = ent->len; + if (len && ent->name[len-1] == '/') + len--; + pos = cache_name_pos(ent->name, len); struct cache_entry *ce; if (0 <= pos) - die("bug in wt_status_print_untracked"); + continue; pos = -pos - 1; if (pos < active_nr) { ce = active_cache[pos]; But the right solution is to refactor this so the code isn't duplicated. And I'll post a patch for that in a second. I do have to wonder, though, whether an even better solution would be to more explicitly tell read_directory "I'm interested only in 'other' files" rather than relying on guessing based on dir.show_other_directories. Then we could just avoid ever passing these gitlinks back to ls-files and status in the first place. -Peff