[PATCH 0/2] dropping stdin support from test-terminal

[PATCH 0/2] dropping stdin support from test-terminal

[Jeff King]

This is an alternative to Rubén's patch here:

  https://lore.kernel.org/git/d95180fc-8f8a-4e1d-987d-3aa0811be7de@gmail.com/

and instead just rips out the stdin feature of test-terminal completely.
There are two tests in t4153 that rely on it (and for which it was
originally added), but I think the fact that we need to use
test_terminal there indicates missing functionality in git-am.

So patch 1 fixes that, and then patch 2 simplifies test-terminal.

  [1/2]: am: add explicit "--retry" option
  [2/2]: test-terminal: drop stdin handling

 Documentation/git-am.txt           |  8 +++++++-
 builtin/am.c                       |  3 +++
 t/t4153-am-resume-override-opts.sh | 14 +++++++++-----
 t/test-terminal.perl               | 29 +++--------------------------
 4 files changed, 22 insertions(+), 32 deletions(-)

-Peff

[PATCH 1/2] am: add explicit "--retry" option

[Jeff King]

After a patch fails, you can ask "git am" to try applying it again with
new options by running without any of the resume options. E.g.:

  git am <patch
  # oops, it failed; let's try again
  git am --3way

But since this second command has no explicit resume option (like
"--continue"), it looks just like an invocation to read a fresh patch
from stdin. To avoid confusing the two cases, there are some heuristics,
courtesy of 8d18550318 (builtin-am: reject patches when there's a
session in progress, 2015-08-04):

	if (in_progress) {
		/*
		 * Catch user error to feed us patches when there is a session
		 * in progress:
		 *
		 * 1. mbox path(s) are provided on the command-line.
		 * 2. stdin is not a tty: the user is trying to feed us a patch
		 *    from standard input. This is somewhat unreliable -- stdin
		 *    could be /dev/null for example and the caller did not
		 *    intend to feed us a patch but wanted to continue
		 *    unattended.
		 */
		if (argc || (resume_mode == RESUME_FALSE && !isatty(0)))
			die(_("previous rebase directory %s still exists but mbox given."),
				state.dir);

		if (resume_mode == RESUME_FALSE)
			resume_mode = RESUME_APPLY;
		[...]

So if no resume command is given, then we require that stdin be a tty,
and otherwise complain about (potentially) receiving an mbox on stdin.
But of course you might not actually have a terminal available! And
sadly there is no explicit way to hit this same code path; this is the
only place that sets RESUME_APPLY. So you're stuck, and scripts like our
test suite have to bend over backwards to create a pseudo-tty.

Let's provide an explicit option to trigger this mode. The code turns
out to be quite simple; just setting "resume_mode" to RESUME_FALSE is
enough to dodge the tty check, and then our state is the same as it
would be with the heuristic case (which we'll continue to allow).

When we don't have a session in progress, there's already code to
complain when resume_mode is set (but we'll add a new test to cover
that).

To test the new option, we'll convert the existing tests that rely on
the fake stdin tty. That lets us test them on more platforms, and will
let us simplify test_terminal a bit in a future patch.

It does, however, mean we're not testing the tty heuristic at all.

Signed-off-by: Jeff King <peff@peff.net>
---
I think even without the test-terminal cleanup, this is a good thing.
Any time there is a heuristic like isatty(), we should have a way for
the user to be more explicit about what they want().

Do we are about losing the testing of the tty heuristic?

 Documentation/git-am.txt           |  8 +++++++-
 builtin/am.c                       |  3 +++
 t/t4153-am-resume-override-opts.sh | 14 +++++++++-----
 3 files changed, 19 insertions(+), 6 deletions(-)

diff --git a/Documentation/git-am.txt b/Documentation/git-am.txt
index 624a6e6fe4..69d5cc9f21 100644
--- a/Documentation/git-am.txt
+++ b/Documentation/git-am.txt
@@ -18,7 +18,7 @@ SYNOPSIS
 	 [--quoted-cr=<action>]
 	 [--empty=(stop|drop|keep)]
 	 [(<mbox> | <Maildir>)...]
-'git am' (--continue | --skip | --abort | --quit | --show-current-patch[=(diff|raw)] | --allow-empty)
+'git am' (--continue | --skip | --abort | --quit | --retry | --show-current-patch[=(diff|raw)] | --allow-empty)
 
 DESCRIPTION
 -----------
@@ -208,6 +208,12 @@ Valid <action> for the `--whitespace` option are:
 	Abort the patching operation but keep HEAD and the index
 	untouched.
 
+--retry::
+	Try to apply the last conflicting patch again. This is generally
+	only useful for passing extra options to the retry attempt
+	(e.g., `--3way`), since otherwise you'll just see the same
+	failure again.
+
 --show-current-patch[=(diff|raw)]::
 	Show the message at which `git am` has stopped due to
 	conflicts.  If `raw` is specified, show the raw contents of
diff --git a/builtin/am.c b/builtin/am.c
index 36839029d2..926592691a 100644
--- a/builtin/am.c
+++ b/builtin/am.c
@@ -2393,6 +2393,9 @@ int cmd_am(int argc, const char **argv, const char *prefix)
 		  N_("show the patch being applied"),
 		  PARSE_OPT_CMDMODE | PARSE_OPT_OPTARG | PARSE_OPT_NONEG | PARSE_OPT_LITERAL_ARGHELP,
 		  parse_opt_show_current_patch, RESUME_SHOW_PATCH_RAW },
+		OPT_CMDMODE(0, "retry", &resume_mode,
+			N_("try to apply current patch again"),
+			RESUME_APPLY),
 		OPT_CMDMODE(0, "allow-empty", &resume_mode,
 			N_("record the empty patch as an empty commit"),
 			RESUME_ALLOW_EMPTY),
diff --git a/t/t4153-am-resume-override-opts.sh b/t/t4153-am-resume-override-opts.sh
index 4add7c7757..a32cec42aa 100755
--- a/t/t4153-am-resume-override-opts.sh
+++ b/t/t4153-am-resume-override-opts.sh
@@ -3,7 +3,6 @@
 test_description='git-am command-line options override saved options'
 
 . ./test-lib.sh
-. "$TEST_DIRECTORY"/lib-terminal.sh
 
 format_patch () {
 	git format-patch --stdout -1 "$1" >"$1".eml
@@ -27,7 +26,12 @@ test_expect_success 'setup' '
 	format_patch side2
 '
 
-test_expect_success TTY '--3way overrides --no-3way' '
+test_expect_success '--retry fails without in-progress operation' '
+	test_must_fail git am --retry 2>err &&
+	test_grep "operation not in progress" err
+'
+
+test_expect_success '--3way overrides --no-3way' '
 	rm -fr .git/rebase-apply &&
 	git reset --hard &&
 	git checkout renamed-file &&
@@ -40,7 +44,7 @@ test_expect_success TTY '--3way overrides --no-3way' '
 
 	# Applying side1 with am --3way will succeed due to the threeway-merge.
 	# Applying side2 will fail as --3way does not apply to it.
-	test_must_fail test_terminal git am --3way </dev/zero &&
+	test_must_fail git am --retry --3way &&
 	test_path_is_dir .git/rebase-apply &&
 	test side1 = "$(cat file2)"
 '
@@ -84,7 +88,7 @@ test_expect_success '--signoff overrides --no-signoff' '
 	test $(git cat-file commit HEAD | grep -c "Signed-off-by:") -eq 0
 '
 
-test_expect_success TTY '--reject overrides --no-reject' '
+test_expect_success '--reject overrides --no-reject' '
 	rm -fr .git/rebase-apply &&
 	git reset --hard &&
 	git checkout first &&
@@ -94,7 +98,7 @@ test_expect_success TTY '--reject overrides --no-reject' '
 	test_path_is_dir .git/rebase-apply &&
 	test_path_is_missing file.rej &&
 
-	test_must_fail test_terminal git am --reject </dev/zero &&
+	test_must_fail git am --retry --reject </dev/zero &&
 	test_path_is_dir .git/rebase-apply &&
 	test_path_is_file file.rej
 '
-- 
2.45.2.817.g6f0d0f2a6c

[PATCH 2/2] test-terminal: drop stdin handling

[Jeff King]

Since 18d8c26930 (test_terminal: redirect child process' stdin to a pty,
2015-08-04), we set up a pty and copy stdin to the child program. But
this ends up being racy; once we send all of the bytes and close the
descriptor, the child program will no longer see a terminal! isatty()
will return 0, and trying to read may return EIO, even if we didn't yet
get all of the bytes.

This was mentioned even in the commit message of 18d8c26930, but we
hacked around it by just sending an infinite input from /dev/zero (in
the intended case, we only cared about isatty(0), not reading actual
input).

And it came up again recently in:

  https://lore.kernel.org/git/d42a55b1-1ba9-4cfb-9c3d-98ea4d86da33@gmail.com/

where we tried to actually send bytes, but they don't always all come
through. So this interface is somewhat of an accident waiting to happen;
a caller might not even care about stdin being a tty, but will get bit
by the flaky behavior.

One solution would probably be to avoid closing test_terminal's end of
the pty altogether. But then the other side would never see EOF on its
stdin.  That may be OK for some cases, but it's another gotcha that
might cause races or deadlocks, depending on what the child expects to
read.

Let's instead just drop test_terminal's stdin feature completely. Since
the previous commit dropped the two cases from t4153 for which the
feature was originally added, there are no callers left that need it.

Signed-off-by: Jeff King <peff@peff.net>
---
 t/test-terminal.perl | 29 +++--------------------------
 1 file changed, 3 insertions(+), 26 deletions(-)

diff --git a/t/test-terminal.perl b/t/test-terminal.perl
index 3810e9bb43..b8fd6a4f13 100755
--- a/t/test-terminal.perl
+++ b/t/test-terminal.perl
@@ -5,17 +5,15 @@
 use IO::Pty;
 use File::Copy;
 
-# Run @$argv in the background with stdio redirected to $in, $out and $err.
+# Run @$argv in the background with stdio redirected to $out and $err.
 sub start_child {
-	my ($argv, $in, $out, $err) = @_;
+	my ($argv, $out, $err) = @_;
 	my $pid = fork;
 	if (not defined $pid) {
 		die "fork failed: $!"
 	} elsif ($pid == 0) {
-		open STDIN, "<&", $in;
 		open STDOUT, ">&", $out;
 		open STDERR, ">&", $err;
-		close $in;
 		close $out;
 		exec(@$argv) or die "cannot exec '$argv->[0]': $!"
 	}
@@ -51,17 +49,6 @@ sub xsendfile {
 	copy($in, $out, 4096) or $!{EIO} or die "cannot copy from child: $!";
 }
 
-sub copy_stdin {
-	my ($in) = @_;
-	my $pid = fork;
-	if (!$pid) {
-		xsendfile($in, \*STDIN);
-		exit 0;
-	}
-	close($in);
-	return $pid;
-}
-
 sub copy_stdio {
 	my ($out, $err) = @_;
 	my $pid = fork;
@@ -81,25 +68,15 @@ sub copy_stdio {
 	die "usage: test-terminal program args";
 }
 $ENV{TERM} = 'vt100';
-my $parent_in = new IO::Pty;
 my $parent_out = new IO::Pty;
 my $parent_err = new IO::Pty;
-$parent_in->set_raw();
 $parent_out->set_raw();
 $parent_err->set_raw();
-$parent_in->slave->set_raw();
 $parent_out->slave->set_raw();
 $parent_err->slave->set_raw();
-my $pid = start_child(\@ARGV, $parent_in->slave, $parent_out->slave, $parent_err->slave);
-close $parent_in->slave;
+my $pid = start_child(\@ARGV, $parent_out->slave, $parent_err->slave);
 close $parent_out->slave;
 close $parent_err->slave;
-my $in_pid = copy_stdin($parent_in);
 copy_stdio($parent_out, $parent_err);
 my $ret = finish_child($pid);
-# If the child process terminates before our copy_stdin() process is able to
-# write all of its data to $parent_in, the copy_stdin() process could stall.
-# Send SIGTERM to it to ensure it terminates.
-kill 'TERM', $in_pid;
-finish_child($in_pid);
 exit($ret);
-- 
2.45.2.817.g6f0d0f2a6c

Re: [PATCH 2/2] test-terminal: drop stdin handling

[Rubén Justo]

On 6/6/24 10:22 AM, Jeff King wrote:
> Since 18d8c26930 (test_terminal: redirect child process' stdin to a pty,
> 2015-08-04), we set up a pty and copy stdin to the child program. But
> this ends up being racy; once we send all of the bytes and close the
> descriptor, the child program will no longer see a terminal! isatty()
> will return 0, and trying to read may return EIO, even if we didn't yet
> get all of the bytes.
> 
> This was mentioned even in the commit message of 18d8c26930, but we
> hacked around it by just sending an infinite input from /dev/zero (in
> the intended case, we only cared about isatty(0), not reading actual
> input).
> 
> And it came up again recently in:
> 
>   https://lore.kernel.org/git/d42a55b1-1ba9-4cfb-9c3d-98ea4d86da33@gmail.com/
> 
> where we tried to actually send bytes, but they don't always all come
> through. So this interface is somewhat of an accident waiting to happen;
> a caller might not even care about stdin being a tty, but will get bit
> by the flaky behavior.
> 
> One solution would probably be to avoid closing test_terminal's end of
> the pty altogether. But then the other side would never see EOF on its
> stdin.  That may be OK for some cases, but it's another gotcha that
> might cause races or deadlocks, depending on what the child expects to
> read.
> 
> Let's instead just drop test_terminal's stdin feature completely. Since
> the previous commit dropped the two cases from t4153 for which the
> feature was originally added, there are no callers left that need it.
> 
> Signed-off-by: Jeff King <peff@peff.net>

Acked-by: Rubén Justo <rjusto@gmail.com>

Re: [PATCH 0/2] dropping stdin support from test-terminal

[Rubén Justo]

On Thu, Jun 06, 2024 at 04:17:24AM -0400, Jeff King wrote:
> This is an alternative to Rubén's patch here:
> 
>   https://lore.kernel.org/git/d95180fc-8f8a-4e1d-987d-3aa0811be7de@gmail.com/
> 
> and instead just rips out the stdin feature of test-terminal completely.

I agree that this is a better approach.

I have no problem building my other series on this.

Thanks!

> There are two tests in t4153 that rely on it (and for which it was
> originally added), but I think the fact that we need to use
> test_terminal there indicates missing functionality in git-am.
> 
> So patch 1 fixes that, and then patch 2 simplifies test-terminal.
> 
>   [1/2]: am: add explicit "--retry" option
>   [2/2]: test-terminal: drop stdin handling
> 
>  Documentation/git-am.txt           |  8 +++++++-
>  builtin/am.c                       |  3 +++
>  t/t4153-am-resume-override-opts.sh | 14 +++++++++-----
>  t/test-terminal.perl               | 29 +++--------------------------
>  4 files changed, 22 insertions(+), 32 deletions(-)
> 
> -Peff

It has been far less than ten, laps ;-)

Re: [PATCH 1/2] am: add explicit "--retry" option

[Junio C Hamano]

Jeff King <peff@peff.net> writes:

> I think even without the test-terminal cleanup, this is a good thing.
> Any time there is a heuristic like isatty(), we should have a way for
> the user to be more explicit about what they want().

I very often do "git am --no-3" to countermand a failed "git am -3"
(or vice versa), so I'll be hit very hard with a need to retrain my
fingers.  But I'll live ;-)

"--retry" is a horrible word, in that it makes it sound like it will
keep trying to apply the same patch over and over until it applies
cleanly or something.  Can't we use "--continue" like everybody else
(like "git rebase --continue", etc.), or would that be even more
confusing?

Re: [PATCH 1/2] am: add explicit "--retry" option

[Junio C Hamano]

Junio C Hamano <gitster@pobox.com> writes:

> Jeff King <peff@peff.net> writes:
>
>> I think even without the test-terminal cleanup, this is a good thing.
>> Any time there is a heuristic like isatty(), we should have a way for
>> the user to be more explicit about what they want().
>
> I very often do "git am --no-3" to countermand a failed "git am -3"
> (or vice versa), so I'll be hit very hard with a need to retrain my
> fingers.  But I'll live ;-)

Ah, no, this is not about not paying attention to isatty(0), but
give us an additional way.  I can see how it would help our tests;
it would be nicer if the feature also has real world use.

But I no longer mind the option existing.

> "--retry" is a horrible word, in that it makes it sound like it will
> keep trying to apply the same patch over and over until it applies
> cleanly or something.  Can't we use "--continue" like everybody else
> (like "git rebase --continue", etc.), or would that be even more
> confusing?

Re: [PATCH 1/2] am: add explicit "--retry" option

[Jeff King]

On Thu, Jun 06, 2024 at 09:48:52AM -0700, Junio C Hamano wrote:

> Junio C Hamano <gitster@pobox.com> writes:
> 
> > Jeff King <peff@peff.net> writes:
> >
> >> I think even without the test-terminal cleanup, this is a good thing.
> >> Any time there is a heuristic like isatty(), we should have a way for
> >> the user to be more explicit about what they want().
> >
> > I very often do "git am --no-3" to countermand a failed "git am -3"
> > (or vice versa), so I'll be hit very hard with a need to retrain my
> > fingers.  But I'll live ;-)
> 
> Ah, no, this is not about not paying attention to isatty(0), but
> give us an additional way.  I can see how it would help our tests;
> it would be nicer if the feature also has real world use.

Exactly, you can still do "git am -3" as before, and that's what I'd
expect everyone to do. It is just about letting you be explicit if you
want.

I don't know if it could have real world use or not. In theory if you
had a more complex program driving "git am", you'd need this. But in
practice, I think the overlap between "people who write GUIs for Git"
and "people who think mailing patches is a good idea" is pretty small.
Let alone one with advanced features like "try this patch again with
--3way". ;)

But I do think as a general rule we should never provide any action
_only_ through heuristics like "is stdin a tty". We should let the user
be explicit, and use heuristics to guess the right thing when they don't
feel like being so.

> > "--retry" is a horrible word, in that it makes it sound like it will
> > keep trying to apply the same patch over and over until it applies
> > cleanly or something.  Can't we use "--continue" like everybody else
> > (like "git rebase --continue", etc.), or would that be even more
> > confusing?

There is already "am --continue", but it is not quite the same thing. It
will try to commit the resolved tree state and keep going. Whereas with
--retry we really are trying the same patch again. So it really is "over
and over again", but only once per invocation. ;)

I'm open to a better name if you can come up with one.

-Peff

Re: [PATCH 1/2] am: add explicit "--retry" option

[Phillip Wood]

Hi Peff and Junio

On 08/06/2024 12:29, Jeff King wrote:
> On Thu, Jun 06, 2024 at 09:48:52AM -0700, Junio C Hamano wrote:
>> Junio C Hamano <gitster@pobox.com> writes:
>>> Jeff King <peff@peff.net> writes:
> But I do think as a general rule we should never provide any action
> _only_ through heuristics like "is stdin a tty". We should let the user
> be explicit, and use heuristics to guess the right thing when they don't
> feel like being so.

I agree that's a good general rule to have

>>> "--retry" is a horrible word, in that it makes it sound like it will
>>> keep trying to apply the same patch over and over until it applies
>>> cleanly or something.

"retry" means "try once more", so I think it is a reasonable name for 
this option. (Programs often retrying a failing operation in a loop but 
then they are retrying many times.)

I haven't looked at these patches in detail so I cannot comment on the 
code changes but I do think the general aim is a good one.

Best Wishes

Phillip