Skip to contentSkip to conversation

git stash apply usability issues

PATCH In review Johannes Sixt 5 participants 8 messages Jun 15, 2016

lore
Share & tools
lore
#0 lore raw

(1) Looking at git-stash.sh I see a few uses of 'git diff' in apply_stash(). Shouldn't these use one of git-diff-{tree,index,files)? The reason is that porcelain 'git diff' invokes custom diff drivers (that in my case run a UI program), whereas the plumbing does not.

Is there a particular reason to use porcelain 'git diff'?

(2) when 'git stash apply' runs merge-recursive, it treats the current state as 'ours' and the stash as 'theirs'. IMHO it should be the other way round: I have stashed away changes to a binary file. Then committed a different modification to it, and now want to apply the stash. This results in a conflict that leaves the current state in the working tree, but I had preferred that the stashed binary file were in the working tree now.

What do other git-stash users think about changing the order?

-- Hannes

#1 lore raw
Quoted 8 lines · Johannes Sixt

(2) when 'git stash apply' runs merge-recursive, it treats the current state as 'ours' and the stash as 'theirs'. IMHO it should be the other way
round: I have stashed away changes to a binary file. Then committed a different modification to it, and now want to apply the stash. This results in a conflict that leaves the current state in the working tree, but I had preferred that the stashed binary file were in the working tree now.

What do other git-stash users think about changing the order?

The current order is the same order that git-rebase uses. I'm not saying its correct, just that its the same as rebase. I think rebase is also backwards and if we change git-stash we should also change git-rebase at the same time (though probably not in the same commit).

Signature · 1 line
Shawn.
#4 lore raw

PATCH 1/1 Avoid invoking diff drivers during git-stash

In reply to Johannes Sixt

git-stash needs to restrict itself to plumbing when running automated diffs as part of its operation as the user may have configured a custom diff driver that opens an interactive UI for certain/all files. Doing that during scripted actions is very unfriendly to the end-user and may cause git-stash to fail to work.

Reported by Johannes Sixt

---

Quoted 6 lines · Johannes Sixt

(1) Looking at git-stash.sh I see a few uses of 'git diff' in apply_stash(). Shouldn't these use one of git-diff-{tree,index,files)? The reason is that porcelain 'git diff' invokes custom diff drivers (that in my case run a UI program), whereas the plumbing does not.

Is there a particular reason to use porcelain 'git diff'?

Does this fix the problem?

 git-stash.sh |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)
git-stash.sh +3 -3
index 7ba6162..def3163 100755
@@ -110,7 +110,7 @@ show_stash () {
 
 	w_commit=$(git rev-parse --verify "$s") &&
 	b_commit=$(git rev-parse --verify "$s^") &&
-	git diff $flags $b_commit $w_commit
+	git diff-tree $flags $b_commit $w_commit
 }
 
 apply_stash () {
@@ -139,7 +139,7 @@ apply_stash () {
 	unstashed_index_tree=
 	if test -n "$unstash_index" && test "$b_tree" != "$i_tree"
 	then
-		git diff --binary $s^2^..$s^2 | git apply --cached
+		git diff-tree --binary $s^2^..$s^2 | git apply --cached
 		test $? -ne 0 &&
 			die 'Conflicts in index. Try without --index.'
 		unstashed_index_tree=$(git-write-tree) ||
@@ -162,7 +162,7 @@ apply_stash () {
 			git read-tree "$unstashed_index_tree"
 		else
 			a="$TMP-added" &&
-			git diff --cached --name-only --diff-filter=A $c_tree >"$a" &&
+			git diff-index --cached --name-only --diff-filter=A $c_tree >"$a" &&
 			git read-tree --reset $c_tree &&
 			git update-index --add --stdin <"$a" ||
 				die "Cannot unstage modified files"
Signature · 1 line
1.5.3.4.1249.g895be
#7 lore raw
In reply to Johannes Sixt
Quoted 3 lines · Johannes Sixt

(2) when 'git stash apply' runs merge-recursive, it treats the current state as 'ours' and the stash as 'theirs'. IMHO it should be the other way round: I have stashed away changes to a binary file. Then committed a different modification to it, and now want to apply the stash. This results in a conflict that leaves the current state in the working tree, but I had preferred that the stashed binary file were in the working tree now.

What do other git-stash users think about changing the order?

Seems right to me. I'd expect to get the stashed version in the working tree in that case.

-Steve

#2 lore raw
In reply to Shawn O. Pearce
Quoted 10 lines · Shawn O. Pearce
Quoted 8 lines · Johannes Sixt

(2) when 'git stash apply' runs merge-recursive, it treats the current state as 'ours' and the stash as 'theirs'. IMHO it should be the other way round: I have stashed away changes to a binary file. Then committed a different modification to it, and now want to apply the stash. This results in a conflict that leaves the current state in the working tree, but I had preferred that the stashed binary file were in the working tree now.

Quoted 4 lines
Quoted 1 line

What do other git-stash users think about changing the order?

The current order is the same order that git-rebase uses. I'm not saying its correct, just that its the same as rebase.

FWIW, StGit push works the same way. The idea being that the current HEAD is our current state ("ours"), and the patch we're pushing is some change we want to apply ("theirs"). I always felt that this was a very natural order of things. But I guess the philosophy in the "stash" case is subtly different, so maybe the change is warranted there.

Signature · 2 lines
Karl Hasselström, [email protected]
      www.treskal.com/kalle
#3 lore raw
Quoted 17 lines · Karl Hasselström
Quoted 10 lines · Shawn O. Pearce
Quoted 8 lines · Johannes Sixt

(2) when 'git stash apply' runs merge-recursive, it treats the current state as 'ours' and the stash as 'theirs'. IMHO it should be the other way round: I have stashed away changes to a binary file. Then committed a different modification to it, and now want to apply the stash. This results in a conflict that leaves the current state in the working tree, but I had preferred that the stashed binary file were in the working tree now.

Quoted 4 lines
Quoted 1 line

What do other git-stash users think about changing the order?

The current order is the same order that git-rebase uses. I'm not saying its correct, just that its the same as rebase.

Quoted 6 lines

FWIW, StGit push works the same way. The idea being that the current HEAD is our current state ("ours"), and the patch we're pushing is some change we want to apply ("theirs"). I always felt that this was a very natural order of things. But I guess the philosophy in the "stash" case is subtly different, so maybe the change is warranted there.

Well, maybe one should then just name this "current" and "separate" instead of "ours" and "theirs".

Signature · 1 line
David Kastrup
#5 lore raw

Re: [PATCH] Avoid invoking diff drivers during git-stash

In reply to Shawn O. Pearce

Shawn,

thanks for the fast response with a patch.

Quoted 9 lines · Shawn O. Pearce
Quoted 6 lines · Johannes Sixt

(1) Looking at git-stash.sh I see a few uses of 'git diff' in apply_stash(). Shouldn't these use one of git-diff-{tree,index,files)? The reason is that porcelain 'git diff' invokes custom diff drivers (that in my case run a UI program), whereas the plumbing does not.

Is there a particular reason to use porcelain 'git diff'?

Does this fix the problem?

It does!

git-stash.sh show_stash () { · line 113

Quoted 6 lines
git-stash.sh +1 -1
@@ -110,7 +110,7 @@ show_stash () {
 
 	w_commit=$(git rev-parse --verify "$s") &&
 	b_commit=$(git rev-parse --verify "$s^") &&
-	git diff $flags $b_commit $w_commit
+	git diff-tree $flags $b_commit $w_commit

However, this porcelain 'git diff' should actually remain because it's part of show_stash().

-- Hannes

#6 lore raw

Re: [PATCH] Avoid invoking diff drivers during git-stash

Quoted 16 lines · Johannes Sixt
Quoted 10 lines · Shawn O. Pearce
Quoted 2 lines · Johannes Sixt

(1) Looking at git-stash.sh I see a few uses of 'git diff' in apply_stash(). Shouldn't these use one of git-diff-{tree,index,files)?

The

Quoted 1 line

reason is that porcelain 'git diff' invokes custom diff drivers (that

in my > case run a UI program), whereas the plumbing does not.

Quoted 2 lines

Is there a particular reason to use porcelain 'git diff'?

Does this fix the problem?

It does!

Quoted 2 lines
(quoted diff) +0 -0
@@ -110,7 +110,7 @@ show_stash () {
 
Quoted 7 lines

git-stash.sh show_stash () { · line 113

Quoted 4 lines
git-stash.sh +1 -1
 	w_commit=$(git rev-parse --verify "$s") &&
 	b_commit=$(git rev-parse --verify "$s^") &&
-	git diff $flags $b_commit $w_commit
+	git diff-tree $flags $b_commit $w_commit

However, this porcelain 'git diff' should actually remain because it's part of show_stash().

Heh. Damn. I was just starting to prepare my evening push and this patch is in maint, which I just merged to master, and I just rebased all of my pu topic branches over that. Junio's Meta toolkit doesn't have an "unrebase" so I can go back and amend that damn commit before pushing.

I'm feeling lazy and don't want to create an unRB right now. I'll probably just throw another commit into maint to fix the above hunk. Thanks for catching it.

Signature · 1 line
Shawn.