threads / patch / 66540

patchpretty: fix lost indent after newline in %w()

Subject: [PATCH] pretty: fix lost indent after newline in %w()

## tl;dr

The pretty-format wrapping code dropped one column of indentation when a hard line break was followed by whitespace. Read the story.

replies: 5people: 3as markdown or json

Andy Koppe· Oct 14, 2026, 17:23 UTC · lore

When strbuf_add_wrapped_text() meets a newline followed by a non-alphanumeric character, it treats it as a hard line break and jumps to the new_line label. That code was written for the case where a line gets too long. It moves past the space where the line was broken by setting the next line's start to 'space + isspace(*space)'.

After a hard line break, 'space' already points to the first character of the next line. If that character is whitespace, it was wrongly skipped, so the line lost one column of indentation. For example,

  git log --format="%w(11) This is%n  the %s%n  commit."

printed " the ..." and " commit." rather than " the ..." and " commit.".

Fix this by skipping the space only in the line-too-long case. In the double-newline case, skip the second newline explicitly, because that case relied on the old skipping.

Add a test for this, and update the expected output of the "magical wrapping directives" test in t4205. There, %+d emits a newline followed by " (tag: ...)", and that leading space is no longer lost. With a wrap width of 1, the space now causes an extra line break.

Signed-off-by: Andy Koppe <andy.koppe@gmail.com>
---
 t/t4202-log.sh                | 6 ++++++
 t/t4205-log-pretty-formats.sh | 2 +-
 utf8.c                        | 4 +++-
 3 files changed, 10 insertions(+), 2 deletions(-)
Show changes to 3 files +10 −2

t/t4202-log.sh, t/t4205-log-pretty-formats.sh, utf8.c

diff --git a/t/t4202-log.sh b/t/t4202-log.sh
index 06552a6271..5895780e8f 100755
--- a/t/t4202-log.sh
+++ b/t/t4202-log.sh
@@ -94,6 +94,12 @@ test_expect_success 'format %w(,1,2)' '
 	test_cmp expect actual
 '
 
+test_expect_success 'format %w(11)' '
+
+	git log -2 --format="%w(11) This is%n  the %s%n  commit." > actual &&
+	test_cmp expect actual
+'
+
 cat > expect << EOF
 $(git rev-parse --short :/sixth  ) sixth
 $(git rev-parse --short :/fifth  ) fifth
diff --git a/t/t4205-log-pretty-formats.sh b/t/t4205-log-pretty-formats.sh
index 4be5c51489..0733f22d58 100755
--- a/t/t4205-log-pretty-formats.sh
+++ b/t/t4205-log-pretty-formats.sh
@@ -1139,7 +1139,7 @@ test_expect_success 'log --pretty with invalid padding format' '
 test_expect_success 'log --pretty with magical wrapping directives' '
 	commit_id=$(git commit-tree HEAD^{tree} -m "describe me") &&
 	git tag describe-me $commit_id &&
-	printf "\n(tag:\ndescribe-me)%%+w(2)" >expect &&
+	printf "\n\n(tag:\ndescribe-me)%%+w(2)" >expect &&
 	git log -1 --pretty="format:%w(1)%+d%+w(2)" $commit_id >actual &&
 	test_cmp expect actual
 '
diff --git a/utf8.c b/utf8.c
index 96460cc414..bb55bf357f 100644
--- a/utf8.c
+++ b/utf8.c
@@ -322,6 +322,7 @@ void strbuf_add_wrapped_text(struct strbuf *buf,
 					space++;
 					if (*space == '\n') {
 						strbuf_addch(buf, '\n');
+						space++;
 						goto new_line;
 					}
 					else if (!isalnum(*space))
@@ -333,9 +334,10 @@ void strbuf_add_wrapped_text(struct strbuf *buf,
 				text++;
 			}
 			else {
+				space += isspace(*space);
 new_line:
 				strbuf_addch(buf, '\n');
-				text = bol = space + isspace(*space);
+				text = bol = space;
 				space = NULL;
 				w = indent = indent2;
 			}
-- 
2.56.0
Andy Koppe· Oct 15, 2026, 20:11 UTC · re: Andy Koppe · lore

Re: [PATCH] pretty: fix lost indent after newline in %w()

Andy Koppe wrote:
Show 8 quoted lines
> After a hard line break, 'space' already points to the first character
> of the next line. If that character is whitespace, it was wrongly
> skipped, so the line lost one column of indentation. For example,
>
>   git log --format="%w(11) This is%n  the %s%n  commit."
>
> printed " the ..." and " commit." rather than "  the ..." and
> "  commit.".

Admittedly, that's a bit of a silly example and test, just to keep it small. My actual use case is "%w(72)%B", because I often have to deal with repos that don't stick to line length standards, which can make a visual mess with --graph or with wide terminal windows.

As a better example, lists of files and their changes like this ...
- foo.c
  Frobble.
- bar.c
  Bobble.
... lose a space of indentation, thus breaking alignment:
- foo.c
 Frobble.
- bar.c
 Bobble.
René Scharfe· Oct 15, 2026, 20:42 UTC · re: Andy Koppe · lore

Re: [PATCH] pretty: fix lost indent after newline in %w()

On 10/14/26 7:23 PM, Andy Koppe wrote:
> When strbuf_add_wrapped_text() meets a newline followed by a
> non-alphanumeric character, it treats it as a hard line break and jumps
> to the new_line label. 

I assume you're referring to ae0b270230 (print_wrapped_text(): allow hard newlines, 2007-11-11), which started to replace single newlines preceding alphanumeric characters with a space, and to keep double newlines and calling them hard.

E.g. the following format pairs are equivalent:
   - hard newline:                    "%w(9)%n%nA" == "%n%nA"
   - replace newline before alpha:    "%w(9)%nA"   == " A"
   - keep newline before punctuation: "%w(9)%n-A"  == "%n-A"
Show 12 quoted lines
> That code was written for the case where a line
> gets too long. It moves past the space where the line was broken by
> setting the next line's start to 'space + isspace(*space)'.
> 
> After a hard line break, 'space' already points to the first character
> of the next line. If that character is whitespace, it was wrongly
> skipped, so the line lost one column of indentation. For example,
> 
>   git log --format="%w(11) This is%n  the %s%n  commit."
> 
> printed " the ..." and " commit." rather than "  the ..." and
> "  commit.".

So currently a newline before whitespace is kept, but that whitespace gets eaten, and you want to keep it instead:

  - now:   "%w(9)%n A" == "%nA"
  - patch: "%w(9)%n A" == "%n A"

The commit message of ae0b270230 doesn't mention this case specifically. I also get the impression that this consumption of the first whitespace character was not intended.

Show 8 quoted lines
> Fix this by skipping the space only in the line-too-long case. In the
> double-newline case, skip the second newline explicitly, because that
> case relied on the old skipping.
> 
> Add a test for this, and update the expected output of the "magical
> wrapping directives" test in t4205. There, %+d emits a newline followed
> by " (tag: ...)", and that leading space is no longer lost. With a wrap
> width of 1, the space now causes an extra line break.

This behavior stood for more than 18 years. Someone might be relying on it -- this test is one example. And you can easily work around it by adding one more space to the format.

On the other hand this hard newline concept is not explained in the docs as far as I can see. That makes it fair game for change in my book (within reason), as long as that doc gap is filled. My 2ct..

René
Junio C Hamano· Oct 15, 2026, 20:59 UTC · re: René Scharfe · lore

Re: [PATCH] pretty: fix lost indent after newline in %w()

René Scharfe <l.s.r@web.de> writes:
> On the other hand this hard newline concept is not explained in the
> docs as far as I can see.  That makes it fair game for change in my
> book (within reason), as long as that doc gap is filled.  My 2ct..

Yeah, to me, it sounds more productive use of our time and a good direction.

Thanks.
Andy Koppe· Oct 17, 2026, 08:26 UTC · re: René Scharfe · lore

Re: [PATCH] pretty: fix lost indent after newline in %w()

On Thu, 15 Oct 2026 at 21:42, René Scharfe <l.s.r@web.de> wrote:
Show 34 quoted lines
> On 10/14/26 7:23 PM, Andy Koppe wrote:
> > When strbuf_add_wrapped_text() meets a newline followed by a
> > non-alphanumeric character, it treats it as a hard line break and jumps
> > to the new_line label.
>
> I assume you're referring to ae0b270230 (print_wrapped_text(): allow
> hard newlines, 2007-11-11), which started to replace single newlines
> preceding alphanumeric characters with a space, and to keep double
> newlines and calling them hard.
>
> E.g. the following format pairs are equivalent:
>
>    - hard newline:                    "%w(9)%n%nA" == "%n%nA"
>    - replace newline before alpha:    "%w(9)%nA"   == " A"
>    - keep newline before punctuation: "%w(9)%n-A"  == "%n-A"
>
> > That code was written for the case where a line
> > gets too long. It moves past the space where the line was broken by
> > setting the next line's start to 'space + isspace(*space)'.
> >
> > After a hard line break, 'space' already points to the first character
> > of the next line. If that character is whitespace, it was wrongly
> > skipped, so the line lost one column of indentation. For example,
> >
> >   git log --format="%w(11) This is%n  the %s%n  commit."
> >
> > printed " the ..." and " commit." rather than "  the ..." and
> > "  commit.".
>
> So currently a newline before whitespace is kept, but that whitespace
> gets eaten, and you want to keep it instead:
>
>   - now:   "%w(9)%n A" == "%nA"
>   - patch: "%w(9)%n A" == "%n A"
Correct.
> The commit message of ae0b270230 doesn't mention this case specifically.
> I also get the impression that this consumption of the first whitespace
> character was not intended.
I agree.
Show 12 quoted lines
> > Fix this by skipping the space only in the line-too-long case. In the
> > double-newline case, skip the second newline explicitly, because that
> > case relied on the old skipping.
> >
> > Add a test for this, and update the expected output of the "magical
> > wrapping directives" test in t4205. There, %+d emits a newline followed
> > by " (tag: ...)", and that leading space is no longer lost. With a wrap
> > width of 1, the space now causes an extra line break.
>
> This behavior stood for more than 18 years.  Someone might be relying
> on it -- this test is one example.  And you can easily work around it
> by adding one more space to the format.

True, but that isn't the actual use case. In v2 I've amended the commit message: "This affected things like code examples or multi-line list items starting with "- " and continuing with two spaces of indentation."

> On the other hand this hard newline concept is not explained in the
> docs as far as I can see.  That makes it fair game for change in my
> book (within reason), as long as that doc gap is filled.  My 2ct..
Very good point. I've added dedicated documentation for %w in v2.

Btw, the detection of paragraph continuations by looking for an alphanumeric ASCII character at the start of the next line could be improved by looking for quotes and opening parentheses too and by supporting Unicode. But I haven't attempted that here.

Andy
Andy Koppe· Oct 17, 2026, 08:11 UTC · re: Andy Koppe · lore

[PATCH v2] pretty: fix lost indent after newline in %w()

When strbuf_add_wrapped_text() meets a newline followed by a non-alphanumeric character, it treats it as a hard line break and jumps to the new_line label. That code was written for the case where a line gets too long. It moves past the space where the line was broken by setting the next line's start to 'space + isspace(*space)'.

After a hard line break, 'space' already points to the first character of the next line. If that character is whitespace, it was wrongly skipped, so the line lost one column of indentation. This affected things like code examples or multi-line list items starting with "- " and continuing with two spaces of indentation.

Fix this by skipping the space only in the line-too-long case. In the double-newline case, skip the second newline explicitly, because that case relied on the old skipping.

Add a test for this, and update the expected output of the "magical wrapping directives" test in t4205. There, %+d emits a newline followed by " (tag: ...)", and that leading space is no longer lost. With a wrap width of 1, the space now causes an extra line break.

Document the %w placeholder properly, removing the reference to `git shortlog -w`, which has different defaults and isn't concerned with paragraph reflowing.

Signed-off-by: Andy Koppe <andy.koppe@gmail.com>
---
 Documentation/pretty-formats.adoc | 9 +++++++--
 t/t4202-log.sh                    | 6 ++++++
 t/t4205-log-pretty-formats.sh     | 2 +-
 utf8.c                            | 4 +++-
 4 files changed, 17 insertions(+), 4 deletions(-)
Show changes to 4 files +16 −4

Documentation/pretty-formats.adoc, t/t4202-log.sh, t/t4205-log-pretty-formats.sh, utf8.c

diff --git a/Documentation/pretty-formats.adoc b/Documentation/pretty-formats.adoc
index 2ae0eb11a9..4fb0419715 100644
--- a/Documentation/pretty-formats.adoc
+++ b/Documentation/pretty-formats.adoc
@@ -169,8 +169,13 @@ The placeholders are:
 	    on the next placeholders until the color is switched
 	    again.
 ++%m++:: left (`<`), right (`>`) or boundary (`-`) mark
-++%w(++`[<w>[,<i1>[,<i2>]]]`++)++:: switch line wrapping, like the `-w` option of
-			    linkgit:git-shortlog[1].
+++%w(++`[<w>[,<i1>[,<i2>]]]`++)++:: set line wrapping and indentation.
+	    Output is word-wrapped at _<w>_ columns, with the first line
+	    indented by _<i1>_ spaces and subsequent lines indented by
+	    _<i2>_ spaces. A width of 0 disables wrapping. Omitted values
+	    default to 0, so ++%w()++ turns off both wrapping and indentation.
+	    Newlines that are followed by an ASCII letter or digit are replaced
+	    by spaces to reflow paragraphs. Existing indentation is kept.
 ++%<(++`<n>[,(trunc|ltrunc|mtrunc)]`++)++:: make the next placeholder take at
 				  least N column widths, padding spaces on
 				  the right if necessary.  Optionally
diff --git a/t/t4202-log.sh b/t/t4202-log.sh
index 06552a6271..5895780e8f 100755
--- a/t/t4202-log.sh
+++ b/t/t4202-log.sh
@@ -94,6 +94,12 @@ test_expect_success 'format %w(,1,2)' '
 	test_cmp expect actual
 '
 
+test_expect_success 'format %w(11)' '
+
+	git log -2 --format="%w(11) This is%n  the %s%n  commit." > actual &&
+	test_cmp expect actual
+'
+
 cat > expect << EOF
 $(git rev-parse --short :/sixth  ) sixth
 $(git rev-parse --short :/fifth  ) fifth
diff --git a/t/t4205-log-pretty-formats.sh b/t/t4205-log-pretty-formats.sh
index 4be5c51489..0733f22d58 100755
--- a/t/t4205-log-pretty-formats.sh
+++ b/t/t4205-log-pretty-formats.sh
@@ -1139,7 +1139,7 @@ test_expect_success 'log --pretty with invalid padding format' '
 test_expect_success 'log --pretty with magical wrapping directives' '
 	commit_id=$(git commit-tree HEAD^{tree} -m "describe me") &&
 	git tag describe-me $commit_id &&
-	printf "\n(tag:\ndescribe-me)%%+w(2)" >expect &&
+	printf "\n\n(tag:\ndescribe-me)%%+w(2)" >expect &&
 	git log -1 --pretty="format:%w(1)%+d%+w(2)" $commit_id >actual &&
 	test_cmp expect actual
 '
diff --git a/utf8.c b/utf8.c
index 96460cc414..bb55bf357f 100644
--- a/utf8.c
+++ b/utf8.c
@@ -322,6 +322,7 @@ void strbuf_add_wrapped_text(struct strbuf *buf,
 					space++;
 					if (*space == '\n') {
 						strbuf_addch(buf, '\n');
+						space++;
 						goto new_line;
 					}
 					else if (!isalnum(*space))
@@ -333,9 +334,10 @@ void strbuf_add_wrapped_text(struct strbuf *buf,
 				text++;
 			}
 			else {
+				space += isspace(*space);
 new_line:
 				strbuf_addch(buf, '\n');
-				text = bol = space + isspace(*space);
+				text = bol = space;
 				space = NULL;
 				w = indent = indent2;
 			}
-- 
2.56.0

← back to recent threads