Skip to contentSkip to conversation

diff: fix regression with --stat and unmerged file

PATCH In review Peter Grayson 2 participants 3 messages Dec 14, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 diff: fix regression with --stat and unmerged file

A regression was introduced in

12fc4ad89e (diff.c: use utf8_strwidth() to count display width, 2022-09-14)

that causes missing newlines after "Unmerged" entries in git diff --cached --stat output.

This problem affects v2.39.0-rc0 through v2.39.0.

This patch adds the missing newline along with a new test to cover this behavior.

---
 diff.c                   |  2 +-
 t/t4046-diff-unmerged.sh | 10 ++++++++++
 2 files changed, 11 insertions(+), 1 deletion(-)
diff.c +1 -1
index 1054a4b732..85f035a9e8 100644
@@ -2801,7 +2801,7 @@ static void show_stats(struct diffstat_t *data, struct diff_options *options)
 		else if (file->is_unmerged) {
 			strbuf_addf(&out, " %s%s%*s | %*s",
 				    prefix, name, padding, "",
-				    number_width, "Unmerged");
+				    number_width, "Unmerged\n");
 			emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
 					 out.buf, out.len, 0);
 			strbuf_reset(&out);
t/t4046-diff-unmerged.sh +10 -0
index 0ae0cd3a52..ffaf69335f 100755
@@ -86,4 +86,14 @@ test_expect_success 'diff-files -3' '
 	test_cmp diff-files-3.expect diff-files-3.actual
 '
 
+test_expect_success 'diff --stat' '
+	for path in $paths
+	do
+		echo " $path | Unmerged" || return 1
+	done >diff-stat.expect &&
+	echo " 0 files changed" >>diff-stat.expect &&
+	git diff --cached --stat >diff-stat.actual &&
+	test_cmp diff-stat.expect diff-stat.actual
+'
+
 test_done
Signature · 1 line
2.39.0
#1 lore raw
Quoted 6 lines · Peter Grayson

A regression was introduced in

12fc4ad89e (diff.c: use utf8_strwidth() to count display width, 2022-09-14)

that causes missing newlines after "Unmerged" entries in git diff --cached --stat output.

Oof, clearly we don't have good test coverage here. Thanks for catching it.

Quoted 2 lines

This patch adds the missing newline along with a new test to cover this behavior.

Both look good to me, but two quick comments:

diff.c static void show_stats(struct diffstat_t *data, struct diff_options *options) · line 2807

Quoted 13 lines
diff.c +1 -1
index 1054a4b732..85f035a9e8 100644
@@ -2801,7 +2801,7 @@ static void show_stats(struct diffstat_t *data, struct diff_options *options)
 		else if (file->is_unmerged) {
 			strbuf_addf(&out, " %s%s%*s | %*s",
 				    prefix, name, padding, "",
-				    number_width, "Unmerged");
+				    number_width, "Unmerged\n");
 			emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
 					 out.buf, out.len, 0);
 			strbuf_reset(&out);

Looking at the offending patch, it also touches "Bin". But that one handles its newline separately (since sometimes there is more data on the line). So this is the only spot that needs to be fixed.

Quoted 7 lines
t/t4046-diff-unmerged.sh +0 -0
index 0ae0cd3a52..ffaf69335f 100755
@@ -86,4 +86,14 @@ test_expect_success 'diff-files -3' '
 	test_cmp diff-files-3.expect diff-files-3.actual
 '

t/t4046-diff-unmerged.sh test_expect_success 'diff-files -3' ' · line 97

Quoted 10 lines
t/t4046-diff-unmerged.sh +9 -0
 
+test_expect_success 'diff --stat' '
+	for path in $paths
+	do
+		echo " $path | Unmerged" || return 1
+	done >diff-stat.expect &&
+	echo " 0 files changed" >>diff-stat.expect &&
+	git diff --cached --stat >diff-stat.actual &&
+	test_cmp diff-stat.expect diff-stat.actual
+'

The rest of this script uses diff-files, but here we're using "diff --cached". It feels like it would be simple to use diff-files for consistency, but strangely it errors out, complaining that the blob can't be read.

This has to do with the setup for the test, which uses "hash-object" without "-w", meaning our index mentions objects we don't actually have. I'm not sure if this is the test trying to cleverly assert that we don't look at the objects themselves, but regardless it seems weird to me that "diff-files" wants to look at the unmerged entries but "diff --cached" doesn't (it does seem like "diff --cached" is right here; diff-files would produce two lines).

So there may be another bug, but if so I don't think it's a recent one. And we are better off working around it for your regression fix, which we'd hope to see merged quickly.

-Peff

#2 lore raw
Quoted 26 lines · Jeff King
Quoted 6 lines · Peter Grayson

A regression was introduced in

12fc4ad89e (diff.c: use utf8_strwidth() to count display width, 2022-09-14)

that causes missing newlines after "Unmerged" entries in git diff --cached --stat output.

Oof, clearly we don't have good test coverage here. Thanks for catching it.

Quoted 2 lines

This patch adds the missing newline along with a new test to cover this behavior.

Both look good to me, but two quick comments:

Quoted 9 lines
diff.c +0 -1
index 1054a4b732..85f035a9e8 100644
@@ -2801,7 +2801,7 @@ static void show_stats(struct diffstat_t *data, struct diff_options *options)
 		else if (file->is_unmerged) {
 			strbuf_addf(&out, " %s%s%*s | %*s",
 				    prefix, name, padding, "",
-				    number_width, "Unmerged");
Quoted 8 lines

diff.c static void show_stats(struct diffstat_t *data, struct diff_options *options) · line 2807

Quoted 4 lines
diff.c +1 -0
+				    number_width, "Unmerged\n");
 			emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
 					 out.buf, out.len, 0);
 			strbuf_reset(&out);

Looking at the offending patch, it also touches "Bin". But that one handles its newline separately (since sometimes there is more data on the line). So this is the only spot that needs to be fixed.

Agreed.

Quoted 23 lines
Quoted 7 lines
t/t4046-diff-unmerged.sh +0 -0
index 0ae0cd3a52..ffaf69335f 100755
@@ -86,4 +86,14 @@ test_expect_success 'diff-files -3' '
 	test_cmp diff-files-3.expect diff-files-3.actual
 '

t/t4046-diff-unmerged.sh test_expect_success 'diff-files -3' ' · line 97

Quoted 10 lines
t/t4046-diff-unmerged.sh +9 -0
 
+test_expect_success 'diff --stat' '
+	for path in $paths
+	do
+		echo " $path | Unmerged" || return 1
+	done >diff-stat.expect &&
+	echo " 0 files changed" >>diff-stat.expect &&
+	git diff --cached --stat >diff-stat.actual &&
+	test_cmp diff-stat.expect diff-stat.actual
+'

The rest of this script uses diff-files, but here we're using "diff --cached". It feels like it would be simple to use diff-files for consistency, but strangely it errors out, complaining that the blob can't be read.

Quoted 7 lines

This has to do with the setup for the test, which uses "hash-object" without "-w", meaning our index mentions objects we don't actually have. I'm not sure if this is the test trying to cleverly assert that we don't look at the objects themselves, but regardless it seems weird to me that "diff-files" wants to look at the unmerged entries but "diff --cached" doesn't (it does seem like "diff --cached" is right here; diff-files would produce two lines).

I also noticed this when I originally wrote the test using diff-files, but didn't dig deeper. Thanks for taking the next step with that. I figure that the "diff" porcelain is a good vehicle to test this particular behavior even if its a bit inconsistent with other cases in this test module.

Quoted 3 lines

So there may be another bug, but if so I don't think it's a recent one. And we are better off working around it for your regression fix, which we'd hope to see merged quickly.

I would also like to see this repair for the clear-and-present regression merged sooner and independent of exploration of the other potential problem with diff-files.

Thanks for taking the time to review this change. I appreciate it!

Pete