Re: [PATCH] contrib/subtree bugfix: Can't `add` annotated tag

James Denholm [off-list ref] writes:

cmd_add_commit() is passed FETCH_HEAD by cmd_add_repository, which is
then rev-parsed into an object ID. However, if the user is fetching a
tag rather than a branch HEAD, such as by executing:

$ git subtree add -P oldGit https://github.com/git/git.git tags/v1.8.0

The object ID is a tag and is never peeled, and the git commit-tree call
(line 561) slaps us in the face because it doesn't handle tag IDs.

The "rev" (not "revs") seems to be used by more things than the
final commit-tree state.  Are we losing some useful information by
peeling it too early like this patch does?  The reason why we
stopped peeling when writing FETCH_HEAD was because we wanted to
record the fact that we merged a tag (and use the GPG signature if
found in it) when constructing the log message for the merge, and
peeling the tag too early and recording the commit in FETCH_HEAD
would make it impossible to do, and I am wondering if this change is
making the same kind of mistake here.

I see that add_msg does not use anything useful from latest_new, so
with the current state of the code, it does not make that much
difference (except that it says "from commit '$latest_new'", and by
peeling, the fact that the user wanted to use a tag is lost from the
result).

On a side note, if merging a tag without --squash, git merge recognises
that it's a tag and adds a note to the merge commit body. It may be
worth mimicking this when using "subtree merge --squash" or
"subtree add".

Yes, and this change makes such a change harder to implement on top,
I suspect.

Would it be sufficient to do

	git commit-tree $tree $headp -p "$rev^0"

in that "not squashing" codepath instead?