patchrevert/cherry-pick: allow starting from dirty work tree.

There is no reason to forbid a dirty work tree when reverting or cherry-picking a change, as long as the index is clean.

The scripted version used to allow it:

    case "$no_commit" in
    t)
    	# We do not intend to commit immediately.  We just want to
    	# merge the differences in.
    	head=$(git-write-tree) ||
    		die "Your index file is unmerged."
    	;;
    *)
    	head=$(git-rev-parse --verify HEAD) ||
    		die "You do not have a valid HEAD"
    	files=$(git-diff-index --cached --name-only $head) || exit
    	if [ "$files" ]; then
    		die "Dirty index: cannot $me (dirty: $files)"
    	fi
    	;;
    esac

but C rewrite tightened the check, probably by mistake.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
---
 builtin-revert.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

Hi,

On Tue, 13 Nov 2007, Junio C Hamano wrote:

Probably.  Thanks.

While we're at cherry-pick: Two days ago I had to rebase in a dirty working directory. Why? Because one of the submodules was not yet ready to be committed to the superproject.

And you cannot easily stash away a submodule.

Now, my quick and dirty solution was to hack a GIT_IGNORE_SUBMODULES patch: http://repo.or.cz/w/git/mingw/4msysgit.git?a=commitdiff;h=9ef6b6bff4e368934c4262af13b0a309be19ebd4

But the more fundamental question is: should we eventually have a mode in cherry-pick (or for that matter, apply!) which can change the submodule? And if so, how to go about it?

I could imagine that in this case, both apply and cherry-pick should call submodule with a to-be-created subcommand to reset the named submodule to a given commit, given the current commit (to avoid races, just as with update-ref <ref> <newsha1> <oldsha1>). Of course, this subcommand would fail if the submodule's working directory is dirty.

Thoughts?

Ciao, Dscho

Johannes Schindelin wrote:

git stash --recursive (to be written)? ;-)

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

I think that would be handled in exactly the same way as switching between two branches that bind different commits at the submodule path.

Your work tree and index knows commit X is at the path, an operation wants to have commit Y at the path as its result.

What's next?

It does not make much difference what that "an operation" is. It can be apply, cherry-pick, revert, merge, or branch switching. I think they can and should follow the same rules.

The most important rule is not to lose local changes, be they in the index or in the work tree. So "an operation" would error out if the index does not match HEAD at such a path, and the work tree does not match the index.

The implementation of "work tree does not match the index" is different when you are talking about a regular blob vs a submodule. For a blob you know what we do. For a submodule, I would imagine that we would check if both the index matches the HEAD and the work tree matches the index in the submodule (this would go recursively).

And after making sure we won't lose local changes, we switch the submodule directory from commit X to commit Y. Most likely detaching the HEAD with "cd sub && git checkout Y^0" or something like that.

Then probably there will be an option added to loosen that "work tree matches the index" check and uses "-m" option to the checkout in the submodule directory.

More tricky is what to do with a submodule that would disappear, though. For that, we would want .git/subprojects/foo.git/ trick we have discussed in the past or something similar.