[PATCH v3 0/2] diff --no-index: support symlinks and pipes

[PATCH v3 0/2] diff --no-index: support symlinks and pipes

[Dennis Kaarsemaker]

git diff <(command1) <(command2) is less useful than it could be, all it outputs is:

diff --git a/dev/fd/63 b/dev/fd/62
index 9e6542b297..9f7b2c291b 120000
--- a/dev/fd/63
+++ b/dev/fd/62
@@ -1 +1 @@
-pipe:[464811685]
\ No newline at end of file
+pipe:[464811687]
\ No newline at end of file

Normal diff provides arguably better output: the diff of the output of the
commands. This series makes it possible for git diff --no-index to follow
symlinks and read from pipes, mimicking the behaviour of normal diff.

v1: http://public-inbox.org/git/20161111201958.2175-1-dennis@kaarsemaker.net/
v2: http://public-inbox.org/git/20170113102021.6054-1-dennis@kaarsemaker.net/

Changes since v2, prompted by feedback from Junio:

- A --derefence option was added and the default is no longer to dereference
  symlinks.
- Instead of looking at what canon_mode returns, use the original mode of 
  files to override behaviour for pipes.
- Turn the !S_ISREG(...) check into a should_mmap_file_contents helper.

Dennis Kaarsemaker (2):
  diff --no-index: optionally follow symlinks
  diff --no-index: support reading from pipes

 Documentation/diff-options.txt |  9 +++++++
 diff-no-index.c                | 16 ++++++++++---
 diff.c                         | 30 +++++++++++++++++++----
 diff.h                         |  2 +-
 t/t4053-diff-no-index.sh       | 54 ++++++++++++++++++++++++++++++++++++++++++
 t/test-lib.sh                  |  4 ++++
 6 files changed, 107 insertions(+), 8 deletions(-)

-- 
2.12.0-437-g0cc2799

[PATCH v3 1/2] diff --no-index: optionally follow symlinks

[Dennis Kaarsemaker]

Git's diff machinery does not follow symlinks, which makes sense as git
itself also does not, but stores the symlink destination.

In --no-index mode however, it is useful for diff to be able to follow
symlinks, matching the behaviour of ordinary diff. A new --dereference
(name copied from diff) option has been added to enable this behaviour.
--no-dereference can be used to disable it again.

Signed-off-by: Dennis Kaarsemaker <dennis@kaarsemaker.net>
---
 Documentation/diff-options.txt |  9 +++++++++
 diff-no-index.c                |  7 ++++---
 diff.c                         | 12 ++++++++++--
 diff.h                         |  2 +-
 t/t4053-diff-no-index.sh       | 44 ++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 68 insertions(+), 6 deletions(-)

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 2d77a19626..5a9d58b701 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -216,6 +216,15 @@ any of those replacements occurred.
 	commit range.  Defaults to `diff.submodule` or the 'short' format
 	if the config option is unset.
 
+ifdef::git-diff[]
+--dereference::
+--no-dereference::
+	Normally, "git diff --no-index" will compare symlinks by comparing what
+	they point to. The `--dereference` option will make it compare the content
+	of the linked files. The `--no-dereference` option disables an earlier
+	`--dereference`.
+endif::git-diff[]
+
 --color[=<when>]::
 	Show colored diff.
 	`--color` (i.e. without '=<when>') is the same as `--color=always`.
diff --git a/diff-no-index.c b/diff-no-index.c
index f420786039..fe48f32ddd 100644
--- a/diff-no-index.c
+++ b/diff-no-index.c
@@ -40,7 +40,7 @@ static int read_directory_contents(const char *path, struct string_list *list)
  */
 static const char file_from_standard_input[] = "-";
 
-static int get_mode(const char *path, int *mode)
+static int get_mode(const char *path, int *mode, int dereference)
 {
 	struct stat st;
 
@@ -52,7 +52,7 @@ static int get_mode(const char *path, int *mode)
 #endif
 	else if (path == file_from_standard_input)
 		*mode = create_ce_mode(0666);
-	else if (lstat(path, &st))
+	else if (dereference ? stat(path, &st) : lstat(path, &st))
 		return error("Could not access '%s'", path);
 	else
 		*mode = st.st_mode;
@@ -93,7 +93,8 @@ static int queue_diff(struct diff_options *o,
 {
 	int mode1 = 0, mode2 = 0;
 
-	if (get_mode(name1, &mode1) || get_mode(name2, &mode2))
+	if (get_mode(name1, &mode1, DIFF_OPT_TST(o, DEREFERENCE)) ||
+		get_mode(name2, &mode2, DIFF_OPT_TST(o, DEREFERENCE)))
 		return -1;
 
 	if (mode1 && mode2 && S_ISDIR(mode1) != S_ISDIR(mode2)) {
diff --git a/diff.c b/diff.c
index be11e4ef2b..2afecfb939 100644
--- a/diff.c
+++ b/diff.c
@@ -2815,7 +2815,7 @@ int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
 		s->size = xsize_t(st.st_size);
 		if (!s->size)
 			goto empty;
-		if (S_ISLNK(st.st_mode)) {
+		if (S_ISLNK(s->mode)) {
 			struct strbuf sb = STRBUF_INIT;
 
 			if (strbuf_readlink(&sb, s->path, s->size))
@@ -2825,6 +2825,10 @@ int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
 			s->should_free = 1;
 			return 0;
 		}
+		if (S_ISLNK(st.st_mode)) {
+			stat(s->path, &st);
+			s->size = xsize_t(st.st_size);
+		}
 		if (size_only)
 			return 0;
 		if ((flags & CHECK_BINARY) &&
@@ -3884,7 +3888,11 @@ int diff_opt_parse(struct diff_options *options,
 	else if (!strcmp(arg, "--no-follow")) {
 		DIFF_OPT_CLR(options, FOLLOW_RENAMES);
 		DIFF_OPT_CLR(options, DEFAULT_FOLLOW_RENAMES);
-	} else if (!strcmp(arg, "--color"))
+	} else if (!strcmp(arg, "--dereference"))
+		DIFF_OPT_SET(options, DEREFERENCE);
+	else if (!strcmp(arg, "--no-dereference"))
+		DIFF_OPT_CLR(options, DEREFERENCE);
+	else if (!strcmp(arg, "--color"))
 		options->use_color = 1;
 	else if (skip_prefix(arg, "--color=", &arg)) {
 		int value = git_config_colorbool(NULL, arg);
diff --git a/diff.h b/diff.h
index 25ae60d5ff..db33dc67f6 100644
--- a/diff.h
+++ b/diff.h
@@ -69,7 +69,7 @@ typedef struct strbuf *(*diff_prefix_fn_t)(struct diff_options *opt, void *data)
 #define DIFF_OPT_FIND_COPIES_HARDER  (1 <<  6)
 #define DIFF_OPT_FOLLOW_RENAMES      (1 <<  7)
 #define DIFF_OPT_RENAME_EMPTY        (1 <<  8)
-/* (1 <<  9) unused */
+#define DIFF_OPT_DEREFERENCE         (1 <<  9)
 #define DIFF_OPT_HAS_CHANGES         (1 << 10)
 #define DIFF_OPT_QUICK               (1 << 11)
 #define DIFF_OPT_NO_INDEX            (1 << 12)
diff --git a/t/t4053-diff-no-index.sh b/t/t4053-diff-no-index.sh
index 453e6c35eb..8c87bffb34 100755
--- a/t/t4053-diff-no-index.sh
+++ b/t/t4053-diff-no-index.sh
@@ -127,4 +127,48 @@ test_expect_success 'diff --no-index from repo subdir respects config (implicit)
 	test_cmp expect actual.head
 '
 
+test_expect_success SYMLINKS 'diff --no-index does not follows symlinks' '
+	echo a >1 &&
+	echo b >2 &&
+	ln -s 1 3 &&
+	ln -s 2 4 &&
+	cat >expect <<-EOF &&
+		--- a/3
+		+++ b/4
+		@@ -1 +1 @@
+		-1
+		\ No newline at end of file
+		+2
+		\ No newline at end of file
+	EOF
+	test_expect_code 1 git diff --no-index 3 4 | tail -n +3 >actual &&
+	test_cmp expect actual
+'
+
+test_expect_success SYMLINKS 'diff --no-index --dereference does follows symlinks' '
+	cat >expect <<-EOF &&
+		--- a/3
+		+++ b/4
+		@@ -1 +1 @@
+		-a
+		+b
+	EOF
+	test_expect_code 1 git diff --no-index --dereference 3 4 | tail -n +3 >actual &&
+	test_cmp expect actual
+'
+
+test_expect_success SYMLINKS 'diff --no-index --no-dereference does not follow symlinks' '
+	cat >expect <<-EOF &&
+		--- a/3
+		+++ b/4
+		@@ -1 +1 @@
+		-1
+		\ No newline at end of file
+		+2
+		\ No newline at end of file
+	EOF
+	test_expect_code 1 git diff --no-index --no-dereference 3 4 | tail -n +3 > actual &&
+	test_cmp expect actual
+'
+
 test_done
-- 
2.12.0-437-g0cc2799

Re: [PATCH v3 1/2] diff --no-index: optionally follow symlinks

[Junio C Hamano]

Dennis Kaarsemaker <dennis@kaarsemaker.net> writes:

> diff --git a/diff.c b/diff.c
> index be11e4ef2b..2afecfb939 100644
> --- a/diff.c
> +++ b/diff.c
> @@ -2815,7 +2815,7 @@ int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
>  		s->size = xsize_t(st.st_size);
>  		if (!s->size)
>  			goto empty;
> -		if (S_ISLNK(st.st_mode)) {
> +		if (S_ISLNK(s->mode)) {
>  			struct strbuf sb = STRBUF_INIT;
>  
>  			if (strbuf_readlink(&sb, s->path, s->size))
> @@ -2825,6 +2825,10 @@ int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
>  			s->should_free = 1;
>  			return 0;
>  		}
> +		if (S_ISLNK(st.st_mode)) {
> +			stat(s->path, &st);
> +			s->size = xsize_t(st.st_size);

Doesn't this affect --no-index mode?  We never need to do a wasteful
stat() after lstat() and we are penalizing the normal codepath with
this change, no?

> @@ -3884,7 +3888,11 @@ int diff_opt_parse(struct diff_options *options,
>  	else if (!strcmp(arg, "--no-follow")) {
>  		DIFF_OPT_CLR(options, FOLLOW_RENAMES);
>  		DIFF_OPT_CLR(options, DEFAULT_FOLLOW_RENAMES);
> -	} else if (!strcmp(arg, "--color"))
> +	} else if (!strcmp(arg, "--dereference"))
> +		DIFF_OPT_SET(options, DEREFERENCE);
> +	else if (!strcmp(arg, "--no-dereference"))
> +		DIFF_OPT_CLR(options, DEREFERENCE);
> +	else if (!strcmp(arg, "--color"))
>  		options->use_color = 1;

Also shouldn't be some code to detect --[no-]dereference options
given when --no-index is not in effect and error out?  As the patch
title says, this change should be a no-op for normal codepath and
only affect the no-index hack.

Re: [PATCH v3 1/2] diff --no-index: optionally follow symlinks

[Dennis Kaarsemaker]

On Sun, 2017-03-19 at 15:14 -0700, Junio C Hamano wrote:
> Dennis Kaarsemaker <dennis@kaarsemaker.net> writes:
> 
> > diff --git a/diff.c b/diff.c
> > index be11e4ef2b..2afecfb939 100644
> > --- a/diff.c
> > +++ b/diff.c
> > @@ -2815,7 +2815,7 @@ int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
> >  		s->size = xsize_t(st.st_size);
> >  		if (!s->size)
> >  			goto empty;
> > -		if (S_ISLNK(st.st_mode)) {
> > +		if (S_ISLNK(s->mode)) {
> >  			struct strbuf sb = STRBUF_INIT;
> >  
> >  			if (strbuf_readlink(&sb, s->path, s->size))
> > @@ -2825,6 +2825,10 @@ int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
> >  			s->should_free = 1;
> >  			return 0;
> >  		}
> > +		if (S_ISLNK(st.st_mode)) {
> > +			stat(s->path, &st);
> > +			s->size = xsize_t(st.st_size);
> 
> Doesn't this affect --no-index mode?  We never need to do a wasteful
> stat() after lstat() and we are penalizing the normal codepath with
> this change, no?

the S_ISLNK(s->mode) conditional above is for the normal codepath,
which returns early. So the stat I added is only done for symlinks in
no_index mode.

> > @@ -3884,7 +3888,11 @@ int diff_opt_parse(struct diff_options *options,
> >  	else if (!strcmp(arg, "--no-follow")) {
> >  		DIFF_OPT_CLR(options, FOLLOW_RENAMES);
> >  		DIFF_OPT_CLR(options, DEFAULT_FOLLOW_RENAMES);
> > -	} else if (!strcmp(arg, "--color"))
> > +	} else if (!strcmp(arg, "--dereference"))
> > +		DIFF_OPT_SET(options, DEREFERENCE);
> > +	else if (!strcmp(arg, "--no-dereference"))
> > +		DIFF_OPT_CLR(options, DEREFERENCE);
> > +	else if (!strcmp(arg, "--color"))
> >  		options->use_color = 1;
> 
> Also shouldn't be some code to detect --[no-]dereference options
> given when --no-index is not in effect and error out?  As the patch
> title says, this change should be a no-op for normal codepath and
> only affect the no-index hack.

But erroring out isn't a no-op. With the current patch you can do 
--dereference without --no-index and it simply wouldn't affect
anything. 

I don't mind either way, so I'll make it error out.
-- 
Dennis Kaarsemaker
http://www.kaarsemaker.net

[PATCH v3 2/2] diff --no-index: support reading from pipes

[Dennis Kaarsemaker]

diff <(command1) <(command2) provides useful output, let's make it
possible for git to do the same.

Signed-off-by: Dennis Kaarsemaker <dennis@kaarsemaker.net>
---
 diff-no-index.c          |  9 +++++++++
 diff.c                   | 18 ++++++++++++++++--
 t/t4053-diff-no-index.sh | 10 ++++++++++
 t/test-lib.sh            |  4 ++++
 4 files changed, 39 insertions(+), 2 deletions(-)

diff --git a/diff-no-index.c b/diff-no-index.c
index fe48f32ddd..1262a587e5 100644
--- a/diff-no-index.c
+++ b/diff-no-index.c
@@ -83,6 +83,15 @@ static struct diff_filespec *noindex_filespec(const char *name, int mode)
 		name = "/dev/null";
 	s = alloc_filespec(name);
 	fill_filespec(s, null_sha1, 0, mode);
+	/*
+	 * In --no-index mode, we support reading from pipes. canon_mode, called by
+	 * fill_filespec, gets confused by this and thinks we now have subprojects.
+	 * To help the rest of the diff machinery along, we now override what
+	 * canon_mode says. This is done here instead of in canon_mode, because the
+	 * rest of git does not (and should not) support pipes.
+	 */
+	if (S_ISFIFO(mode))
+		s->mode = S_IFREG | ce_permissions(mode);
 	if (name == file_from_standard_input)
 		populate_from_stdin(s);
 	return s;
diff --git a/diff.c b/diff.c
index 2afecfb939..4f74a54d74 100644
--- a/diff.c
+++ b/diff.c
@@ -2765,6 +2765,11 @@ static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
 	return 0;
 }
 
+static int should_mmap_file_contents(struct stat *st)
+{
+	return S_ISREG(st->st_mode);
+}
+
 /*
  * While doing rename detection and pickaxe operation, we may need to
  * grab the data for the blob (or file) for our own in-core comparison.
@@ -2839,9 +2844,18 @@ int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
 		fd = open(s->path, O_RDONLY);
 		if (fd < 0)
 			goto err_empty;
-		s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
+		if (!should_mmap_file_contents(&st)) {
+			struct strbuf sb = STRBUF_INIT;
+			strbuf_read(&sb, fd, 0);
+			s->size = sb.len;
+			s->data = strbuf_detach(&sb, NULL);
+			s->should_free = 1;
+		}
+		else {
+			s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
+			s->should_munmap = 1;
+		}
 		close(fd);
-		s->should_munmap = 1;
 
 		/*
 		 * Convert from working tree format to canonical git format
diff --git a/t/t4053-diff-no-index.sh b/t/t4053-diff-no-index.sh
index 8c87bffb34..2d9b322315 100755
--- a/t/t4053-diff-no-index.sh
+++ b/t/t4053-diff-no-index.sh
@@ -171,4 +171,14 @@ test_expect_success SYMLINKS 'diff --no-index --no-dereference does not follow s
 	test_cmp expect actual
 '
 
+test_expect_success PROCESS_SUBSTITUTION 'diff --no-index works on fifos' '
+	cat >expect <<-EOF &&
+		@@ -1 +1 @@
+		-1
+		+2
+	EOF
+	test_expect_code 1 git diff --no-index --dereference <(echo 1) <(echo 2) | tail -n +5 > actual &&
+	test_cmp expect actual
+'
+
 test_done
diff --git a/t/test-lib.sh b/t/test-lib.sh
index 11562bde10..78f3d24651 100644
--- a/t/test-lib.sh
+++ b/t/test-lib.sh
@@ -1128,3 +1128,7 @@ build_option () {
 test_lazy_prereq LONG_IS_64BIT '
 	test 8 -le "$(build_option sizeof-long)"
 '
+
+test_lazy_prereq PROCESS_SUBSTITUTION '
+	eval "foo=<(echo test)" 2>/dev/null
+'
-- 
2.12.0-437-g0cc2799

Re: [PATCH v3 0/2] diff --no-index: support symlinks and pipes

[Junio C Hamano]

Dennis Kaarsemaker <dennis@kaarsemaker.net> writes:

> Normal diff provides arguably better output: the diff of the output of the
> commands. This series makes it possible for git diff --no-index to follow
> symlinks and read from pipes, mimicking the behaviour of normal diff.
>
> v1: http://public-inbox.org/git/20161111201958.2175-1-dennis@kaarsemaker.net/
> v2: http://public-inbox.org/git/20170113102021.6054-1-dennis@kaarsemaker.net/
>
> Changes since v2, prompted by feedback from Junio:
>
> - A --derefence option was added and the default is no longer to dereference
>   symlinks.

I do agree that it makes sense to have --[no-]dereference options,
but I do not think it was my feedback and suggestion to make it
optional (not default) to dereference, so please do not blame me for
that choice.

Re: [PATCH v3 0/2] diff --no-index: support symlinks and pipes

[Dennis Kaarsemaker]

On Sun, 2017-03-19 at 15:08 -0700, Junio C Hamano wrote:
> Dennis Kaarsemaker <dennis@kaarsemaker.net> writes:
> 
> > Normal diff provides arguably better output: the diff of the output of the
> > commands. This series makes it possible for git diff --no-index to follow
> > symlinks and read from pipes, mimicking the behaviour of normal diff.
> > 
> > v1: http://public-inbox.org/git/20161111201958.2175-1-dennis@kaarsemaker.net/
> > v2: http://public-inbox.org/git/20170113102021.6054-1-dennis@kaarsemaker.net/
> > 
> > Changes since v2, prompted by feedback from Junio:
> > 
> > - A --derefence option was added and the default is no longer to dereference
> >   symlinks.
> 
> I do agree that it makes sense to have --[no-]dereference options,
> but I do not think it was my feedback and suggestion to make it
> optional (not default) to dereference, so please do not blame me for
> that choice.

Then I misinterpreted your message at 
http://public-inbox.org/git/xmqqk29yedkv.fsf@gitster.mtv.corp.google.com/
No blame inteded, my apologies for coming across as blaming.

-- 
Dennis Kaarsemaker
http://www.kaarsemaker.net

Re: [PATCH v3 0/2] diff --no-index: support symlinks and pipes

[Junio C Hamano]

Dennis Kaarsemaker <dennis@kaarsemaker.net> writes:

> On Sun, 2017-03-19 at 15:08 -0700, Junio C Hamano wrote:
> ...
>> > - A --derefence option was added and the default is no longer to dereference
>> >   symlinks.
>> 
>> I do agree that it makes sense to have --[no-]dereference options,
>> but I do not think it was my feedback and suggestion to make it
>> optional (not default) to dereference, so please do not blame me for
>> that choice.
>
> Then I misinterpreted your message at 
> http://public-inbox.org/git/xmqqk29yedkv.fsf@gitster.mtv.corp.google.com/
> No blame inteded, my apologies for coming across as blaming.

s/blame/credit/ then.  I do not too deeply care which one is the
default, and if we were adding --no-index without any existing users
today, I probably would suggest making it deref by default (i.e. to
make "diff --no-index" match better what other peoples' diffs do),
but that would be a behaviour change to existing users if done today,
so I think what you did probably is a good thing.

Thanks.