patchapply: fix new-style empty context line triggering incomplete-line check
7 messages between Mar 17, 2026 and Mar 31, 2026, from Junio C Hamano, Eric Sunshine, D. Ben Knoble.
Plain Markdown or JSON for tools and agents. Diffs are folded; open one to read it.
A new-style unified context diff represents an empty context line with an empty line (instead of a line with a single SP on it). The code to check whitespace errors in an incoming patch is designed to omit the first byte of a line (typically SP, "-", or "+") and pass the remainder of the line to the whitespace checker.
Usually we do not pass a context line to the whitespace error checker, but when we are correcting errors, we do. This "remove the first byte and send the remainder" strategy of checking a line ended up sending a zero-length string to the whitespace checker when seeing a new-style empty context line, which caused the whitespace checker to say "ah, you do not even have a newline at the end!", leading to an "incomplete line" in the middle of the patch!
Fix this by pretending that we got a traditional empty context line when we drive the whitespace checker.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
---
apply.c | 12 ++++++++++--
t/t4124-apply-ws-rule.sh | 16 ++++++++++++++++
2 files changed, 26 insertions(+), 2 deletions(-)
Show changes to 2 files +26 −2
apply.c, t/t4124-apply-ws-rule.sh
diff --git a/apply.c b/apply.c
index f01204d15b..e88e5c77e3 100644
--- a/apply.c
+++ b/apply.c
@@ -1796,8 +1796,16 @@ static int parse_fragment(struct apply_state *state,
trailing++;
check_old_for_crlf(patch, line, len);
if (!state->apply_in_reverse &&
- state->ws_error_action == correct_ws_error)
- check_whitespace(state, line, len, patch->ws_rule);
+ state->ws_error_action == correct_ws_error) {
+ const char *test_line = line;
+ int test_len = len;
+ if (*line == '\n') {
+ test_line = " \n";
+ test_len = 2;
+ }
+ check_whitespace(state, test_line, test_len,
+ patch->ws_rule);
+ }
break;
case '-':
if (!state->apply_in_reverse)
diff --git a/t/t4124-apply-ws-rule.sh b/t/t4124-apply-ws-rule.sh
index 29ea7d4268..8573e12f46 100755
--- a/t/t4124-apply-ws-rule.sh
+++ b/t/t4124-apply-ws-rule.sh
@@ -561,6 +561,22 @@ test_expect_success 'check incomplete lines (setup)' '
git config core.whitespace incomplete-line
'
+test_expect_success 'no incomplete context line (not an error)' '
+ test_when_finished "rm -f sample*-i patch patch-new target" &&
+ (test_write_lines 1 2 3 "" 4 5 ) >sample-i &&
+ (test_write_lines 1 2 3 "" 0 5 ) >sample2-i &&
+ cat sample-i >target &&
+ git add target &&
+ cat sample2-i >target &&
+ git diff-files -p target >patch &&
+ sed -e "s/^ $//" <patch >patch-new &&
+
+ cat sample-i >target &&
+ git apply --whitespace=fix <patch-new 2>error &&
+ test_cmp sample2-i target &&
+ test_must_be_empty error
+'
+
test_expect_success 'incomplete context line (not an error)' '
(test_write_lines 1 2 3 4 5 && printf 6) >sample-i &&
(test_write_lines 1 2 3 0 5 && printf 6) >sample2-i &&
--
2.53.0-769-g8689fa97fd
Re: [PATCH] apply: fix new-style empty context line triggering incomplete-line check
On Tue, Mar 17, 2026 at 2:01 PM Junio C Hamano <gitster@pobox.com> wrote:
Show 28 quoted lines
> A new-style unified context diff represents an empty context line
> with an empty line (instead of a line with a single SP on it). The
> code to check whitespace errors in an incoming patch is designed to
> omit the first byte of a line (typically SP, "-", or "+") and pass the
> remainder of the line to the whitespace checker.
>
> Usually we do not pass a context line to the whitespace error checker,
> but when we are correcting errors, we do. This "remove the first
> byte and send the remainder" strategy of checking a line ended up
> sending a zero-length string to the whitespace checker when seeing a
> new-style empty context line, which caused the whitespace checker to
> say "ah, you do not even have a newline at the end!", leading to an
> "incomplete line" in the middle of the patch!
>
> Fix this by pretending that we got a traditional empty context line
> when we drive the whitespace checker.
>
> Signed-off-by: Junio C Hamano <gitster@pobox.com>
> ---
> diff --git a/t/t4124-apply-ws-rule.sh b/t/t4124-apply-ws-rule.sh
> index 29ea7d4268..8573e12f46 100755
> --- a/t/t4124-apply-ws-rule.sh
> +++ b/t/t4124-apply-ws-rule.sh
> @@ -561,6 +561,22 @@ test_expect_success 'check incomplete lines (setup)' '
> +test_expect_success 'no incomplete context line (not an error)' '
> + test_when_finished "rm -f sample*-i patch patch-new target" &&
> + (test_write_lines 1 2 3 "" 4 5 ) >sample-i &&
> + (test_write_lines 1 2 3 "" 0 5 ) >sample2-i &&
Curious. Why are the `test_write_line` invocations wrapped in parentheses?
Also, is the whitespace before the closing parenthesis intentional?
> test_expect_success 'incomplete context line (not an error)' '
> (test_write_lines 1 2 3 4 5 && printf 6) >sample-i &&
> (test_write_lines 1 2 3 0 5 && printf 6) >sample2-i &&
Perhaps the parentheses in the new test were copied from some existing test, such as this, which already used them for a legitimate reason?
Re: [PATCH] apply: fix new-style empty context line triggering incomplete-line check
Eric Sunshine <sunshine@sunshineco.com> writes:
Show 14 quoted lines
>> + test_when_finished "rm -f sample*-i patch patch-new target" &&
>> + (test_write_lines 1 2 3 "" 4 5 ) >sample-i &&
>> + (test_write_lines 1 2 3 "" 0 5 ) >sample2-i &&
>
> Curious. Why are the `test_write_line` invocations wrapped in parentheses?
>
> Also, is the whitespace before the closing parenthesis intentional?
>
>> test_expect_success 'incomplete context line (not an error)' '
>> (test_write_lines 1 2 3 4 5 && printf 6) >sample-i &&
>> (test_write_lines 1 2 3 0 5 && printf 6) >sample2-i &&
>
> Perhaps the parentheses in the new test were copied from some existing
> test, such as this, which already used them for a legitimate reason?
Yes, the existing one was concatenating output from two commands run in a row into a single redirection, so (grouping of the commands) in parentheses were justifiable.
The new one does not have such a justification. Thanks for noticing.
Re: [PATCH] apply: fix new-style empty context line triggering incomplete-line check
On Tue, Mar 17, 2026 at 2:48 PM Junio C Hamano <gitster@pobox.com> wrote:
Show 24 quoted lines
>
> Eric Sunshine <sunshine@sunshineco.com> writes:
>
> >> + test_when_finished "rm -f sample*-i patch patch-new target" &&
> >> + (test_write_lines 1 2 3 "" 4 5 ) >sample-i &&
> >> + (test_write_lines 1 2 3 "" 0 5 ) >sample2-i &&
> >
> > Curious. Why are the `test_write_line` invocations wrapped in parentheses?
> >
> > Also, is the whitespace before the closing parenthesis intentional?
> >
> >> test_expect_success 'incomplete context line (not an error)' '
> >> (test_write_lines 1 2 3 4 5 && printf 6) >sample-i &&
> >> (test_write_lines 1 2 3 0 5 && printf 6) >sample2-i &&
> >
> > Perhaps the parentheses in the new test were copied from some existing
> > test, such as this, which already used them for a legitimate reason?
>
> Yes, the existing one was concatenating output from two commands run
> in a row into a single redirection, so (grouping of the commands) in
> parentheses were justifiable.
>
> The new one does not have such a justification. Thanks for
> noticing.
I think braces { test_writes lines … && printf … ; } would have sufficed for the second example, and might be cheaper (avoiding the extra process for the subshell, which we've been told is especially expensive for Windows).
--
D. Ben Knoble
Re: [PATCH] apply: fix new-style empty context line triggering incomplete-line check
On Wed, Mar 18, 2026 at 12:36 PM D. Ben Knoble <ben.knoble@gmail.com> wrote:
Show 20 quoted lines
> On Tue, Mar 17, 2026 at 2:48 PM Junio C Hamano <gitster@pobox.com> wrote:
> > Eric Sunshine <sunshine@sunshineco.com> writes:
> > >> test_expect_success 'incomplete context line (not an error)' '
> > >> (test_write_lines 1 2 3 4 5 && printf 6) >sample-i &&
> > >> (test_write_lines 1 2 3 0 5 && printf 6) >sample2-i &&
> > >
> > > Perhaps the parentheses in the new test were copied from some existing
> > > test, such as this, which already used them for a legitimate reason?
> >
> > Yes, the existing one was concatenating output from two commands run
> > in a row into a single redirection, so (grouping of the commands) in
> > parentheses were justifiable.
> >
> > The new one does not have such a justification. Thanks for
> > noticing.
>
> I think braces { test_writes lines … && printf … ; } would have
> sufficed for the second example, and might be cheaper (avoiding the
> extra process for the subshell, which we've been told is especially
> expensive for Windows).No doubt. I considered making the exact same comment, however, this is existing code which Junio's patch was not touching, so the comment would not have been directly applicable except possibly as a #leftoverbits for someone to tackle as a mini-project or some such, so I opted against saying anything about it.
Re: [PATCH] apply: fix new-style empty context line triggering incomplete-line check
"D. Ben Knoble" <ben.knoble@gmail.com> writes:
> I think braces { test_writes lines … && printf … ; } would have
> sufficed for the second exampleNobody complained about them ever since they were written. More importantly, we have plenty of them that nobody complained and bothered to uglify so far [*].
;-)
$ git grep -e '(.*) >' master -- t/t[0-9]\*.sh | grep -v -e '\$(' -e 'cd ' | wc -l
60
$ git grep -e '(.*) >' master -- t/t[0-9]\*.sh | grep -v -e '\$(' -e 'cd ' |
head -n 20
master:t/t3050-subprojects-fetch.sh: (git rev-parse HEAD && git ls-files -s) >expected &&
master:t/t3050-subprojects-fetch.sh: (git rev-parse HEAD && git ls-files -s) >../actual
master:t/t3050-subprojects-fetch.sh: (git rev-parse HEAD && git ls-files -s) >expected &&
master:t/t3050-subprojects-fetch.sh: (git rev-parse HEAD && git ls-files -s) >../actual
master:t/t3402-rebase-merge.sh: (echo "0 $T" && cat original) >renamed &&
master:t/t3900-i18n-commit.sh: (sed "1,/^$/d" raw | iconv -f $new -t utf-8) >actual &&
master:t/t4001-diff-rename.sh: (cat path1 && echo new) >new-path &&
master:t/t4015-diff-whitespace.sh: (echo foo && echo baz | tr -d "\012") >x &&
master:t/t4015-diff-whitespace.sh: (echo bar && echo baz | tr -d "\012") >x &&
master:t/t4019-diff-wserror.sh:if (grep "$blue_grep" <check-grep | grep "$blue_grep") >/dev/null 2>&1
master:t/t4019-diff-wserror.sh:elif (grep -a "$blue_grep" <check-grep | grep -a "$blue_grep") >/dev/null 2>&1
master:t/t4024-diff-optimize-common.sh: ( zs $n && echo a ) >file-a$n &&
master:t/t4024-diff-optimize-common.sh: ( echo b && zs $n && echo ) >file-b$n &&
master:t/t4024-diff-optimize-common.sh: ( printf c && zs $n ) >file-c$n &&
master:t/t4024-diff-optimize-common.sh: ( echo d && zs $n ) >file-d$n &&
master:t/t4024-diff-optimize-common.sh: ( zs $n && echo A ) >file-a$n &&
master:t/t4024-diff-optimize-common.sh: ( echo B && zs $n && echo ) >file-b$n &&
master:t/t4024-diff-optimize-common.sh: ( printf C && zs $n ) >file-c$n &&
master:t/t4024-diff-optimize-common.sh: ( echo D && zs $n ) >file-d$n &&
master:t/t4101-apply-nonl.sh:(echo a; echo b) >frotz.0[Footnote]
* I personally find that we need ';' immediately before '}'
intolerably ugly.
Re: [PATCH] apply: fix new-style empty context line triggering incomplete-line check
Junio C Hamano <gitster@pobox.com> writes:
Show 22 quoted lines
> A new-style unified context diff represents an empty context line
> with an empty line (instead of a line with a single SP on it). The
> code to check whitespace errors in an incoming patch is designed to
> omit the first byte of a line (typically SP, "-", or "+") and pass the
> remainder of the line to the whitespace checker.
>
> Usually we do not pass a context line to the whitespace error checker,
> but when we are correcting errors, we do. This "remove the first
> byte and send the remainder" strategy of checking a line ended up
> sending a zero-length string to the whitespace checker when seeing a
> new-style empty context line, which caused the whitespace checker to
> say "ah, you do not even have a newline at the end!", leading to an
> "incomplete line" in the middle of the patch!
>
> Fix this by pretending that we got a traditional empty context line
> when we drive the whitespace checker.
>
> Signed-off-by: Junio C Hamano <gitster@pobox.com>
> ---
> apply.c | 12 ++++++++++--
> t/t4124-apply-ws-rule.sh | 16 ++++++++++++++++
> 2 files changed, 26 insertions(+), 2 deletions(-)
There were only comments on the unnecessary uses of subshell in test, which were fixed since then, and I've been using them in production without problems, so let me mark this for 'next' now.
Objections and better yet polishing on top are of course welcome.
Show 50 quoted lines
> diff --git a/apply.c b/apply.c
> index f01204d15b..e88e5c77e3 100644
> --- a/apply.c
> +++ b/apply.c
> @@ -1796,8 +1796,16 @@ static int parse_fragment(struct apply_state *state,
> trailing++;
> check_old_for_crlf(patch, line, len);
> if (!state->apply_in_reverse &&
> - state->ws_error_action == correct_ws_error)
> - check_whitespace(state, line, len, patch->ws_rule);
> + state->ws_error_action == correct_ws_error) {
> + const char *test_line = line;
> + int test_len = len;
> + if (*line == '\n') {
> + test_line = " \n";
> + test_len = 2;
> + }
> + check_whitespace(state, test_line, test_len,
> + patch->ws_rule);
> + }
> break;
> case '-':
> if (!state->apply_in_reverse)
> diff --git a/t/t4124-apply-ws-rule.sh b/t/t4124-apply-ws-rule.sh
> index 29ea7d4268..8573e12f46 100755
> --- a/t/t4124-apply-ws-rule.sh
> +++ b/t/t4124-apply-ws-rule.sh
> @@ -561,6 +561,22 @@ test_expect_success 'check incomplete lines (setup)' '
> git config core.whitespace incomplete-line
> '
>
> +test_expect_success 'no incomplete context line (not an error)' '
> + test_when_finished "rm -f sample*-i patch patch-new target" &&
> + (test_write_lines 1 2 3 "" 4 5 ) >sample-i &&
> + (test_write_lines 1 2 3 "" 0 5 ) >sample2-i &&
> + cat sample-i >target &&
> + git add target &&
> + cat sample2-i >target &&
> + git diff-files -p target >patch &&
> + sed -e "s/^ $//" <patch >patch-new &&
> +
> + cat sample-i >target &&
> + git apply --whitespace=fix <patch-new 2>error &&
> + test_cmp sample2-i target &&
> + test_must_be_empty error
> +'
> +
> test_expect_success 'incomplete context line (not an error)' '
> (test_write_lines 1 2 3 4 5 && printf 6) >sample-i &&
> (test_write_lines 1 2 3 0 5 && printf 6) >sample2-i &&