Make git-contacts accept patch contents via stdin for better interoperability with other utilities. Read from stdin under the following circumstances:

* stdin is not a terminal
* argv is empty

Cc: Junio C Hamano <gitster@pobox.com>
Cc: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Brigham Campbell <me@brighamcampbell.com>
---
I authored this patch as the first of a two-patch series to enable b4 to
automatically determine patch recipients. The second patch modifies
.b4-config to wire everything up. I discovered, however, that b4 doesn't
yet pass the mbox "Fixes " header as it should. I sent a fix to b4.

The second patch of this series will therefore have to wait until the b4 fix lands in the latest version, so I'm splitting up this series of patches to git. This patch is useful and warranted even by itself.

Link: https://lore.kernel.org/tools/20260914-mbox-from-auto-to-cc-v1-1-fbf0c3fbd460@brighamcampbell.com
---
 contrib/contacts/git-contacts | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

Brigham Campbell <me@brighamcampbell.com> writes:

It is unusual for these trailers to be here for this project.

Immediately before this hunk, there is

for (@files) {
	scan_patch_file(\%sources, $_);
}

where both @rev_args and @files are derived from @ARGV. In fact, once @ARGV is sifted into @rev_args and @files, nobody uses @ARGV afterwards. @ARGV is used only to decide which things to work on and how, and the decision is recorded in variables with more meaningful names than @ARGV. After we are done with @ARGV, the actions are carried out using variables with more meaningful names.

It would be nicer if you followed the pattern by doing something like the attached patch instead.

By the way, there is a comment at the beginning of the file that says:

    # List people who might be interested in a patch.  Useful as the argument to
    # git-send-email --cc-cmd option, and in other situations.
    #
    # Usage: git contacts <file | rev-list option> ...

We'd probably need to update that part of the file as well.

 contrib/contacts/git-contacts | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

Make git-contacts accept patch contents via stdin for better interoperability with other utilities. Read from stdin under the following circumstances:

* stdin is not a terminal
* argv is empty

Signed-off-by: Brigham Campbell <me@brighamcampbell.com>
---
I authored this patch as the first of a two-patch series to enable b4 to
automatically determine patch recipients. The second patch modifies
.b4-config to wire everything up. I discovered, however, that the latest
version of b4 doesn't pass the mbox "Fixes " header as it should. A fix
is queued for the next release of b4.

The second patch of this series will therefore have to wait until the b4 fix lands in the latest version, so I'm splitting up this series of patches to git. This patch is useful and warranted even by itself.

Link: https://lore.kernel.org/tools/20260914-mbox-from-auto-to-cc-v1-1-fbf0c3fbd460@brighamcampbell.com
---
Changes in v2 (Thanks for your careful review, Junio):
- Minor variable cleanup / un-spaghettification
- Include update to usage comment
- Remove Cc trailers from commit message
- Link to v1: https://patch.msgid.link/20260914-git-contacts-stdin-v1-1-9ac628e6fd20@brighamcampbell.com
---
 contrib/contacts/git-contacts | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

Brigham Campbell <me@brighamcampbell.com> writes:

Not necessarily a suggestion to change what you already added, but a mere food for thought.

An obvious (and UNIX-y) alternative design would have been to follow the popular convention to tell the command to read from the standard input stream when "-" is used instead of a filename. Then this line did not have to change, and the command would have allowed arguments like:

	$ git contacts patch1 patch2 - <patch3
	$ git contacts patch1 - patch3 <patch2

On Wed Sep 16, 2026 at 8:25 AM MDT, Junio C Hamano wrote:

I considered this approach as well. I may as well have flipped a coin. My approach to choosing hyphen or no wasn't very scientific.

Naturally, if you'd prefer that I change the approach to instead allow the user to pass "-" to indicate that a patch should be read via stdin, I'm happy to send out another revision. I understand why we would rather get the interface right on the first try, even for something small like this.

Until you ask me to work a revision that uses "-", I'll await further review on this v2.

Thanks again for your time!

Make git-contacts accept patch contents via stdin for better interoperability with other utilities. Read from stdin when the user passes `-` at least once:

$ git contacts - <patch $ git contacts patch1 - patch3 <patch2

Signed-off-by: Brigham Campbell <me@brighamcampbell.com>
---
This patch is motivated by an ongoing personal effort to allow b4 to
invoke git-contacts automatically.

This latest revision requires the user to pass '-' instead of implicitly
reading from stdin when stdin is not a tty and argv is empty. This
simplifies the change slightly and allows for input via files,
revisions, and stdin in a single invocation of git-contacts.
---
Changes in v3:
- Make user pass '-' instead of an empty argv and non-TTY stdin
- Link to v2: https://patch.msgid.link/20260915-git-contacts-stdin-v2-1-2005061d907a@brighamcampbell.com

Changes in v2:
- Minor variable cleanup / un-spaghettification
- Include update to usage comment
- Remove Cc trailers from commit message
- Link to v1: https://patch.msgid.link/20260914-git-contacts-stdin-v1-1-9ac628e6fd20@brighamcampbell.com
---
 contrib/contacts/git-contacts | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

Brigham Campbell <me@brighamcampbell.com> writes:

The second one is an example that we could tell the command to read these three files in patch1 and then patch2 and then patch3.

But this does not match what I would expect here, which would be to have something like this ...

        if (-e || $_ eq '-') {
	        push @files, $_;
        } else {
		push @rev_args, $_;
	}

... around here, and then tell scan_patch_file that "-" shouldn't be opened and instead read from the standard input stream, perhaps like

        sub scan_patch_file {
                my ($commits, $file) = @_;
                if ($file eq '-') {
                        $f = \*STDIN;
                } else {
                        open my $f, '<', $file or die "read failure: $file: $!\n";
                }
                scan_patches($commits, undef, $f);
                close $f;
        }

or something.

On Wed Sep 23, 2026 at 10:13 PM MDT, Junio C Hamano wrote:

Does the order in which arguments are processed matter? If it's a matter of producing predictable output, I just checked out `master` and I see that running git-contacts twice produces the same email addresses, but in different orders for the same two input patches:

	$ git format-patch HEAD~2
	0001-Git-2.56-rc2.patch
	0002-Revert-Merge-branch-en-no-amend-during-conflicts.patch
	$ ./contrib/contacts/git-contacts *.patch
	Ævar Arnfjörð Bjarmason <avarab@gmail.com>
	Elijah Newren <newren@gmail.com>
	Junio C Hamano <gitster@pobox.com>
	$ ./contrib/contacts/git-contacts *.patch
	Elijah Newren <newren@gmail.com>
	Ævar Arnfjörð Bjarmason <avarab@gmail.com>
	Junio C Hamano <gitster@pobox.com>

What's more, git-contacts already disregards the order of argv. It will always scan file contents first and then git rev-lists second, even if the user passes the rev-list first like follows:

	git contacts HEAD~2.. foo.patch

I would argue that this is not by mistake. The user should have no expectation as to the order in which arguments are processed. If the commit message implies that the arguments are processed one after another, regardless of whether it's a file, a rev-list, or stdin, then the commit message should be changed.

Do I misunderstand?

"Brigham Campbell" <me@brighamcampbell.com> writes:

It does not matter if the order makes difference in the end result, because my comment was made out of principle. I would not at all surprised if the command gathers information from all sources, sorts and then emits---in which case the end result won't be obviously affected.

Make git-contacts accept patch contents via stdin for better interoperability with other utilities. Read from stdin when the user passes `-` at least once:

$ git contacts - <patch

Signed-off-by: Brigham Campbell <me@brighamcampbell.com>
---
Junio, you suggested that I make git-contacts process files and stdin
with respect to the order in which they appear in argv. I instead
removed the implication that git-contacts processes input in any
particular order.

Even before this patch, git-contacts parses files first, then git rev-lists second, regardless of their order in argv. If we instead want git-contacts to parse arguments in the same order that they're passed, that's a change which should occur in a separate patch.

I'm happy to write that patch if you'd like.
---
Changes in v4:
- Don't imply that git-contacts processes input in any particular order
- Link to v3: https://patch.msgid.link/20260923-git-contacts-stdin-v3-1-56dd43c64d56@brighamcampbell.com

Changes in v3:
- Make user pass '-' instead of an empty argv and non-TTY stdin
- Link to v2: https://patch.msgid.link/20260915-git-contacts-stdin-v2-1-2005061d907a@brighamcampbell.com

Changes in v2:
- Minor variable cleanup / un-spaghettification
- Include update to usage comment
- Remove Cc trailers from commit message
- Link to v1: https://patch.msgid.link/20260914-git-contacts-stdin-v1-1-9ac628e6fd20@brighamcampbell.com

To: git@vger.kernel.org
Cc: Junio C Hamano <gitster@pobox.com>
Cc: Patrick Steinhardt <ps@pks.im>
---
 contrib/contacts/git-contacts | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

Brigham Campbell <me@brighamcampbell.com> writes:

OK.

Fair enough.

We would want an update to contrib/contacts/git-contacts.adoc as well. Here is my attempt. I realize that the new text implies that we read only one patch from the standard input, but I do not know offhand whether scan_patches() successfully reads multiple patches concatenated together into a single stream. If it does, we may want to say "read patch file(s) from" instead.

The original uses <rev> in <angle brackets> to clarify that it is a placeholder, so the added text does the same for <patch>. The original text does not say what the command expects of the "patch files" it mentions, which may be worth improving and when it happens we would say <patch> in the description to refer to one argument on the command line. But I left it outside the scope of this message.

 contrib/contacts/git-contacts.adoc | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Make git-contacts accept patches via stdin. Multiple patches may be concatenated together before being passed into git-contacts; git-contacts recognizes the mbox `From ` header inserted by git-format-patch to separate concatenated patches.

Update git-contacts and its corresponding documentation.

---
Changes in v5:
- Add a patch documenting stdin support
- Link to v4: https://patch.msgid.link/20260925-git-contacts-stdin-v4-1-9b4e4bcbb91c@brighamcampbell.com

Changes in v4:
- Don't imply that git-contacts processes input in any particular order
- Link to v3: https://patch.msgid.link/20260923-git-contacts-stdin-v3-1-56dd43c64d56@brighamcampbell.com

Changes in v3:
- Make user pass '-' instead of an empty argv and non-TTY stdin
- Link to v2: https://patch.msgid.link/20260915-git-contacts-stdin-v2-1-2005061d907a@brighamcampbell.com

Changes in v2:
- Minor variable cleanup / un-spaghettification
- Include update to usage comment
- Remove Cc trailers from commit message
- Link to v1: https://patch.msgid.link/20260914-git-contacts-stdin-v1-1-9ac628e6fd20@brighamcampbell.com

To: git@vger.kernel.org
Cc: Junio C Hamano <gitster@pobox.com>
Cc: Patrick Steinhardt <ps@pks.im>
---
Brigham Campbell (2):
      git-contacts: allow inputting patch via stdin
      git-contacts: add stdin functionality to docs

 contrib/contacts/git-contacts      | 9 +++++++--
 contrib/contacts/git-contacts.adoc | 3 ++-
 2 files changed, 9 insertions(+), 3 deletions(-)
---
base-commit: 0f8e75abebff0877cae681a3d5ff31ac47f54220
change-id: 20260914-git-contacts-stdin-1e829930e19b

Thanks!

Make git-contacts accept patch contents via stdin for better interoperability with other utilities. Read from stdin when the user passes `-` at least once:

$ git contacts - <patch

Signed-off-by: Brigham Campbell <me@brighamcampbell.com>
---
 contrib/contacts/git-contacts | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

git-contacts now accepts patches via stdin. Optionally, multiple patches may be concatenated before piping the result into git-contacts. Document this feature.

Signed-off-by: Brigham Campbell <me@brighamcampbell.com>
---
 contrib/contacts/git-contacts.adoc | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Brigham Campbell <me@brighamcampbell.com> writes:

The end result may be the same, but I somehow expected that a new feature plus the documentation update to describe the new feature to come in a single patch.

Thanks.

Brigham Campbell <me@brighamcampbell.com> writes:

Doesn't the Usage comment at the beginning also want to be updated?

Thanks.

On Tue Sep 29, 2026 at 1:29 PM MDT, Junio C Hamano wrote:

I'm inclined to agree with you. I would have updated it had you not commented the following on v2, referring to the Usage comment:

On Wed Sep 16, 2026 at 8:25 AM MDT, Junio C Hamano wrote:

I took that to mean that v5 did not have to change the Usage comment. I will update it in v6.

Thanks for your patience,

Make git-contacts accept patch contents via stdin for better interoperability with other utilities. Read from stdin when the user passes `-` at least once:

$ git contacts - <patch

Signed-off-by: Brigham Campbell <me@brighamcampbell.com>
---
I verified the documentation changes by rendering html and inspecting
the output in a web browser.
---
Changes in v6:
- Squash documentation and code into a single commit
- Make documentation changes more complete
- Link to v5: https://patch.msgid.link/20260928-git-contacts-stdin-v5-0-e9becaebc47e@brighamcampbell.com

Changes in v5:
- Add a patch documenting stdin support
- Link to v4: https://patch.msgid.link/20260925-git-contacts-stdin-v4-1-9b4e4bcbb91c@brighamcampbell.com

Changes in v4:
- Don't imply that git-contacts processes input in any particular order
- Link to v3: https://patch.msgid.link/20260923-git-contacts-stdin-v3-1-56dd43c64d56@brighamcampbell.com

Changes in v3:
- Make user pass '-' instead of an empty argv and non-TTY stdin
- Link to v2: https://patch.msgid.link/20260915-git-contacts-stdin-v2-1-2005061d907a@brighamcampbell.com

Changes in v2:
- Minor variable cleanup / un-spaghettification
- Include update to usage comment
- Remove Cc trailers from commit message
- Link to v1: https://patch.msgid.link/20260914-git-contacts-stdin-v1-1-9ac628e6fd20@brighamcampbell.com

To: git@vger.kernel.org
Cc: Junio C Hamano <gitster@pobox.com>
Cc: Patrick Steinhardt <ps@pks.im>
---
 contrib/contacts/git-contacts      | 11 ++++++++---
 contrib/contacts/git-contacts.adoc | 14 +++++++++++---
 2 files changed, 19 insertions(+), 6 deletions(-)

Brigham Campbell <me@brighamcampbell.com> writes:

Very much appreciated.  Will replace.

Let me mark the topic for 'next'.

Thanks.