Create a new commit from patches without using any worktree?

Hello

Is there any way to record a new commit based on contents in a patch
without going through a worktree? Of course the operation would have
to fail if the patch does not apply cleanly. The problem is I find
myself doing things like this:

put away current work on some branch A
checkout some branch B that is only used for patching
run git patch --check to see if patches apply cleanly. *reject* if
not. apply otherwise.
return to branch A.

I.e. the checkout of B is redundant because I'm anyway not going to
try to resolve any conflicts.

I know I can do something similar to what I need with "git-fetch
src:dst", except it's rather awkward if what you have is patches. I
suspect there must be some way to mimic that fetch behaviour manually,
and in a way that lets me choose exactly what to apply?

BR / Klas

Hi Klas,

Klas Lindberg wrote:

Is there any way to record a new commit based on contents in a patch
without going through a worktree?

Not sure if there a user-friendly way to do this, but if you are
scripting, I'd suggest looking into "git read-tree", "git apply
--cached", "git write-tree", "git commit-tree", and GIT_INDEX_FILE.

Hope that helps,
Jonathan

Am 10/27/2010 3:49, schrieb Klas Lindberg:

Is there any way to record a new commit based on contents in a patch
without going through a worktree?

This script may be a starter. It takes a commit on the command line
instead of a patch, but it processes that commit as a patch ("diff-tree -p"),
so the script should be easy to adapt.

-- Hannes

-- 8< --
#!/bin/sh

OPTIONS_SPEC="\
git post to-ref [from-rev]
--
"
. git-sh-setup

while test $# != 0
do
	case "$1" in
	--)	shift; break;;
	-*)	usage;;
	*)	break;;
	esac
	shift
done

FROM=$(git rev-parse --verify --symbolic-full-name "$1") || exit
shift
if test $# = 0; then
	set -- HEAD
fi
test $# = 1 || usage

# populate a temporary index
tmpidx=$GIT_DIR/index-post-$$
git read-tree --index-output="$tmpidx" "$FROM" || exit
GIT_INDEX_FILE=$tmpidx
export GIT_INDEX_FILE
trap 'rm -f "$tmpidx"' 0 1 2 15

git diff-tree -p -M -C "$@" | git apply --cached || exit

newtree=$(git write-tree) &&
newrev=$(
	eval "$(get_author_ident_from_commit "$1")"
	git-cat-file commit "$1" | sed -e '1,/^$/d' |
	git commit-tree $newtree -p "$FROM"
) || exit

if git check-ref-format "$FROM"
then
	set_reflog_action post
	subject=$(git log --no-walk --pretty=%s "$newrev") &&
	git update-ref -m "$GIT_REFLOG_ACTION: $subject" "$FROM" $newrev || exit
fi
if test -z "$GIT_QUIET"
then
	git rev-list -1 --oneline $newrev
fi

On Wed, Oct 27, 2010 at 11:07, Johannes Sixt [off-list ref] wrote:

Am 10/27/2010 3:49, schrieb Klas Lindberg:

quoted

Is there any way to record a new commit based on contents in a patch
without going through a worktree?

This script may be a starter. It takes a commit on the command line
instead of a patch, but it processes that commit as a patch ("diff-tree -p"),
so the script should be easy to adapt.

-- Hannes

-- 8< --
#!/bin/sh

OPTIONS_SPEC="\
git post to-ref [from-rev]
--
"
. git-sh-setup

while test $# != 0
do
       case "$1" in
       --)     shift; break;;
       -*)     usage;;
       *)      break;;
       esac
       shift
done

FROM=$(git rev-parse --verify --symbolic-full-name "$1") || exit
shift
if test $# = 0; then
       set -- HEAD
fi
test $# = 1 || usage

# populate a temporary index
tmpidx=$GIT_DIR/index-post-$$

I think you should honor GIT_INDEX_FILE here, so that you can
guarantee that tmpidx is on the same device as the index file used by
git read-tree.

Bert

git read-tree --index-output="$tmpidx" "$FROM" || exit
GIT_INDEX_FILE=$tmpidx
export GIT_INDEX_FILE
trap 'rm -f "$tmpidx"' 0 1 2 15

git diff-tree -p -M -C "$@" | git apply --cached || exit

newtree=$(git write-tree) &&
newrev=$(
       eval "$(get_author_ident_from_commit "$1")"
       git-cat-file commit "$1" | sed -e '1,/^$/d' |
       git commit-tree $newtree -p "$FROM"
) || exit

if git check-ref-format "$FROM"
then
       set_reflog_action post
       subject=$(git log --no-walk --pretty=%s "$newrev") &&
       git update-ref -m "$GIT_REFLOG_ACTION: $subject" "$FROM" $newrev || exit
fi
if test -z "$GIT_QUIET"
then
       git rev-list -1 --oneline $newrev
fi
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Ha! It works! Thank you so very very much. I've been trying to figure
this out on my own for a while.

BR / Klas

On Wed, Oct 27, 2010 at 11:07 AM, Johannes Sixt [off-list ref] wrote:

Am 10/27/2010 3:49, schrieb Klas Lindberg:

quoted

Is there any way to record a new commit based on contents in a patch
without going through a worktree?

This script may be a starter. It takes a commit on the command line
instead of a patch, but it processes that commit as a patch ("diff-tree -p"),
so the script should be easy to adapt.

-- Hannes

-- 8< --
#!/bin/sh

OPTIONS_SPEC="\
git post to-ref [from-rev]
--
"
. git-sh-setup

while test $# != 0
do
       case "$1" in
       --)     shift; break;;
       -*)     usage;;
       *)      break;;
       esac
       shift
done

FROM=$(git rev-parse --verify --symbolic-full-name "$1") || exit
shift
if test $# = 0; then
       set -- HEAD
fi
test $# = 1 || usage

# populate a temporary index
tmpidx=$GIT_DIR/index-post-$$
git read-tree --index-output="$tmpidx" "$FROM" || exit
GIT_INDEX_FILE=$tmpidx
export GIT_INDEX_FILE
trap 'rm -f "$tmpidx"' 0 1 2 15

git diff-tree -p -M -C "$@" | git apply --cached || exit

newtree=$(git write-tree) &&
newrev=$(
       eval "$(get_author_ident_from_commit "$1")"
       git-cat-file commit "$1" | sed -e '1,/^$/d' |
       git commit-tree $newtree -p "$FROM"
) || exit

if git check-ref-format "$FROM"
then
       set_reflog_action post
       subject=$(git log --no-walk --pretty=%s "$newrev") &&
       git update-ref -m "$GIT_REFLOG_ACTION: $subject" "$FROM" $newrev || exit
fi
if test -z "$GIT_QUIET"
then
       git rev-list -1 --oneline $newrev
fi