Previously, the working tree state was modified with `patch', which
was a fragile operation.  Do everything with `git-apply --cached
--index' instead.
---
 cg-commit |   33 +++++++++++++++++----------------
 1 files changed, 17 insertions(+), 16 deletions(-)

diff --git a/cg-commit b/cg-commit
index 360d5e6..f676e1f 100755
--- a/cg-commit
+++ b/cg-commit
@@ -575,22 +575,6 @@ else
 fi
 rm "$LOGMSG2"
 
-if [ "$review" ]; then
-	if ! cmp -s "$PATCH" "$PATCH2"; then
-		echo "Reverting the original patch..."
-		if ! cg-patch -R < "$PATCH"; then
-			die "unable to revert the original patch; the original patch is available in $PATCH, your edited patch is available in $PATCH2, your log message is in $LOGMSG, your working copy is in undefined state now and the world is about to end in ten minutes, have a nice day"
-		fi
-		echo "Applying the edited patch..."
-		if ! cg-patch < "$PATCH2"; then
-			# FIXME: Do something better to alleviate this situation.
-			# At least restore the tree to the original state.
-			die "unable to apply the edited patch; the original patch is available in $PATCH, your edited patch is available in $PATCH2, your log message is in $LOGMSG, your working copy is in undefined state now and the world is about to end in five minutes, have a nice day"
-		fi
-	fi
-fi
-
-
 precommit_update()
 {
 	queueN=(); queueD=(); queueM=();
@@ -620,6 +604,23 @@ if [ ! "$ignorecache" ]; then
 	precommit_update "${commitfiles[@]}" || die "update-cache failed"
 fi
 
+if [ "$review" ]; then
+	if ! cmp -s "$PATCH" "$PATCH2"; then
+		git-read-tree HEAD
+		while ! git-apply --check --cached --index "$PATCH2"
+		do
+			echo "patch tried was:"
+			cat $PATCH2
+			echo "your patch does not apply cleanly, re-edit it!"
+			echo -n "sleeping 5s..."
+			sleep 4
+			echo "get it right this time OK?"
+			sleep 1
+			${EDITOR:-vi} "$PATCH2"
+		done
+		git-apply --cached --index "$PATCH2"
+	fi
+fi
 
 oldhead=
 oldheadname="$(git-symbolic-ref HEAD)"
-- ...
[ message continues ]

I do not use Cogito so I do not know what behaviour is wanted
here, but '--cached --index' is same as saying just '--cached'
as far as I know.  It will patch against the index and should
not touch working tree.  If the original used 'patch' to apply,
I suspect it wanted to touch the working tree (and possibly, it
wanted to leave the index alone?), so --cached might be
completely wrong thing to use here?

-

[ message continues ]

The context is that "cg-commit -p", a kind of poor man's interactive
commit that lets you preview changes that are to be committed in 'diff'
form, and edit the patch to be applied. Many users expect this command
to behave this way; they're quite surprised and sometimes even miffed
when the changes they deleted from the patch are gone from their working
copy.

Sam.
-

[ message continues ]

Ah, I see that's why you do want to leave the working tree
untouched.  I think '--cached' alone is the right way to spell
it (strictly speaking, giving --index and --cached should error
out, although the current implementation does not).

-

[ message continues ]