Re: git subtree issue in more recent versions

From: Junio C Hamano <gitster@pobox.com>
Date: Apr 29, 2014, 18:22 UTC
Message-ID: <xmqqa9b4kncm.fsf@gitster.dls.corp.google.com>
In-Reply-To: <55893188F2F68B4B9819D7F9452F981D09FCDF67@NTXBOIMBX01.micron.com>

"Kevin Cagle (kcagle) [CONT - Type 2]" <kcagle@micron.com> writes:

Show 9 quoted lines

> $ git subtree add -P oldGit https://github.com/git/git.git tags/v1.9.2
>
> Will produce this error:
>
> 10ff115f5c572299de4e04ade0d7adb3c75fbf1f is not a valid 'commit' object
>
> The bug isn't found in 1.7.1 (installed subtree manually) but is found in 1.9.0 and 2.0.0.rc1. 
>
> It's related to the git fetch putting the "wrong" SHA1 in .git/FETCH_HEAD.

The change 7a2b128d is very much deliberate; we wanted not to lose information that the user was trying to pull a tag not a commit, because not unwrapping a tag to a commit it points at too early is essential to allow pulling and merging a signed tag, which was released as part of the 1.7.9 that happened in late January 2012 (whew, is it already more than two years ago? time flies).

    commit 7a2b128d13d880635e7317a9208cfa42a660f143
    Author: Linus Torvalds <torvalds@linux-foundation.org>
    Date:   Wed Nov 2 19:19:34 2011 -0700

    fetch: do not store peeled tag object names in FETCH_HEAD
    
    We do not want to record tags as parents of a merge when the user does
    "git pull $there tag v1.0" to merge tagged commit, but that is not a good
    enough excuse to peel the tag down to commit when storing in FETCH_HEAD.
    The caller of underlying "git fetch $there tag v1.0" may have other uses
    of information contained in v1.0 tag in mind.

If the caller of "fetch" (like the codepath in subtree) must see a commit object, it needs to unwrap the tag itself in the new (eh, not so new, though ;-) world order.