[PATCH 0/4] git-related updates

[PATCH 0/4] git-related updates

[Eric Wong]

core.abbrev=no is now used in preparation for SHA-256 support.

I started working on supporting git 2.36+ --batch-command, but having
to support prior versions will require more refactoring, I think...

2/4 was painful :<

Eric Wong (4):
  git: use core.abbrev=no on git 2.31+
  process_pipe: warn hackers off using it for bidirectional pipes
  git: drop needless ENOENT import
  git: reduce delete ops in _destroy

 lib/PublicInbox/Git.pm         | 42 ++++++++++++++++++++++++----------
 lib/PublicInbox/ProcessPipe.pm |  6 +++--
 2 files changed, 34 insertions(+), 14 deletions(-)

[PATCH 1/4] git: use core.abbrev=no on git 2.31+

[Eric Wong]

This makes it easier to support SHA-256 inboxes in the future.
Tested with both git 2.30.2 (Debian stable) and 2.39.1
---
 lib/PublicInbox/Git.pm | 27 +++++++++++++++++++++++----
 1 file changed, 23 insertions(+), 4 deletions(-)

diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index c9ed48be..ff3ac40f 100644
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -16,7 +16,7 @@ use Errno qw(EINTR EAGAIN ENOENT);
 use File::Glob qw(bsd_glob GLOB_NOSORT);
 use File::Spec ();
 use Time::HiRes qw(stat);
-use PublicInbox::Spawn qw(popen_rd spawn);
+use PublicInbox::Spawn qw(popen_rd which);
 use PublicInbox::Tmpfile;
 use IO::Poll qw(POLLIN);
 use Carp qw(croak carp);
@@ -46,7 +46,6 @@ my %GIT_ESC = (
 );
 my %ESC_GIT = map { $GIT_ESC{$_} => $_ } keys %GIT_ESC;
 
-
 # unquote pathnames used by git, see quote.c::unquote_c_style.c in git.git
 sub git_unquote ($) {
 	return $_[0] unless ($_[0] =~ /\A"(.*)"\z/);
@@ -122,6 +121,26 @@ sub _bidi_pipe {
 		}
 		return;
 	}
+
+	state $EXE_ST = ''; # pack('dd', st_ctime, st_size);
+	my $exe = which('git') // die "git not found in $ENV{PATH}";
+	my @st = stat($exe) or die "stat: $!";
+	my $st = pack('dd', $st[10], $st[7]);
+	state $VER;
+	if ($st ne $EXE_ST) {
+		my $rd = popen_rd([ $exe, '--version' ]);
+		my $v = readline($rd);
+		$v =~ /\b([0-9]+(?:\.[0-9]+){2})/ or die
+			"$exe --version output: $v # unparseable";
+		my @v = split(/\./, $1, 3);
+		$VER = ($v[0] << 24) | ($v[1] << 16) | $v[2];
+		$EXE_ST = $st;
+	}
+
+	# git 2.31.0+ supports -c core.abbrev=no, don't bother with
+	# core.abbrev=64 since not many releases had SHA-256 prior to 2.31
+	my $abbr = $VER < (2 << 24 | 31 << 16) ? 40 : 'no';
+
 	pipe(my ($out_r, $out_w)) or $self->fail("pipe failed: $!");
 	my $rdr = { 0 => $out_r, pgid => 0 };
 	my $gd = $self->{git_dir};
@@ -129,8 +148,8 @@ sub _bidi_pipe {
 		$rdr->{-C} = $gd;
 		$gd = $1;
 	}
-	my @cmd = (qw(git), "--git-dir=$gd",
-			qw(-c core.abbrev=40 cat-file), $batch);
+	my @cmd = ($exe, "--git-dir=$gd", '-c', "core.abbrev=$abbr",
+			'cat-file', $batch);
 	if ($err) {
 		my $id = "git.$self->{git_dir}$batch.err";
 		my $fh = tmpfile($id) or $self->fail("tmpfile($id): $!");

[PATCH 2/4] process_pipe: warn hackers off using it for bidirectional pipes

[Eric Wong]

While most uses of ->DESTROY happens in a predictable order in
long-lived daemons, process teardown on exit is chaotic and not
subject to ordering guarantees, so we must keep both ends of a
`git cat-file --batch*' pipe at the same level in the object
hierarchy.

Drop an old Carp import while I'm in the area.
---
 lib/PublicInbox/Git.pm         | 1 +
 lib/PublicInbox/ProcessPipe.pm | 6 ++++--
 2 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index ff3ac40f..a3813bf2 100644
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -156,6 +156,7 @@ sub _bidi_pipe {
 		$self->{$err} = $fh;
 		$rdr->{2} = $fh;
 	}
+	# see lib/PublicInbox/ProcessPipe.pm for why we don't use that here
 	my ($in_r, $p) = popen_rd(\@cmd, undef, $rdr);
 	awaitpid($self->{$pid} = $p, undef);
 	$self->{"$pid.owner"} = $$;
diff --git a/lib/PublicInbox/ProcessPipe.pm b/lib/PublicInbox/ProcessPipe.pm
index 068631c6..1bc792c4 100644
--- a/lib/PublicInbox/ProcessPipe.pm
+++ b/lib/PublicInbox/ProcessPipe.pm
@@ -1,10 +1,12 @@
 # Copyright (C) all contributors <meta@public-inbox.org>
 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
 
-# a tied handle for auto reaping of children tied to a pipe, see perltie(1)
+# a tied handle for auto reaping of children tied to a read-only pipe, see perltie(1)
+# DO NOT use this as-is for bidirectional pipes/sockets (e.g. in PublicInbox::Git),
+# both ends of the pipe must be at the same level of the Perl object hierarchy
+# to ensure orderly destruction.
 package PublicInbox::ProcessPipe;
 use v5.12;
-use Carp qw(carp);
 use PublicInbox::DS qw(awaitpid);
 
 sub waitcb { # awaitpid callback

[PATCH 3/4] git: drop needless ENOENT import

[Eric Wong]

I imported it in commit 356439a571c536eaa487031802b436d087113f4f
(gcf2 + extsearch: check for unlinked files on Linux, 2021-09-22)
but never used it.
---
 lib/PublicInbox/Git.pm | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index a3813bf2..9197ea67 100644
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -12,7 +12,7 @@ use v5.10.1;
 use parent qw(Exporter);
 use POSIX ();
 use IO::Handle; # ->autoflush
-use Errno qw(EINTR EAGAIN ENOENT);
+use Errno qw(EINTR EAGAIN);
 use File::Glob qw(bsd_glob GLOB_NOSORT);
 use File::Spec ();
 use Time::HiRes qw(stat);

[PATCH 4/4] git: reduce delete ops in _destroy

[Eric Wong]

We can avoid some extra returns and branches by just relying on
variadic arguments.
---
 lib/PublicInbox/Git.pm | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/lib/PublicInbox/Git.pm b/lib/PublicInbox/Git.pm
index 9197ea67..12f997dc 100644
--- a/lib/PublicInbox/Git.pm
+++ b/lib/PublicInbox/Git.pm
@@ -373,13 +373,11 @@ sub check {
 }
 
 sub _destroy {
-	my ($self, $rbuf, $in, $out, $pid, $err) = @_;
-	delete @$self{($rbuf, $in, $out)};
-	delete $self->{$err} if $err; # `err_c'
+	my ($self, $pid, @rest) = @_; # rest = rbuf, in, out, err
+	my ($p) = delete @$self{($pid, @rest)};
 
 	# GitAsyncCat::event_step may delete {$pid}
-	my $p = delete($self->{$pid}) // return;
-	awaitpid($p) if $$ == $self->{"$pid.owner"};
+	awaitpid($p) if defined($p) && $$ == $self->{"$pid.owner"};
 }
 
 sub async_abort ($) {
@@ -468,8 +466,8 @@ sub cleanup {
 	async_wait_all($self);
 	delete $self->{inflight};
 	delete $self->{inflight_c};
-	_destroy($self, qw(rbuf in out pid));
-	_destroy($self, qw(rbuf_c in_c out_c pid_c err_c));
+	_destroy($self, qw(pid rbuf in out));
+	_destroy($self, qw(pid_c rbuf_c in_c out_c err_c));
 	undef;
 }