-# Copyright (C) 2014-2020 all contributors <meta@public-inbox.org>
+# Copyright (C) 2014-2021 all contributors <meta@public-inbox.org>
# License: GPLv2 or later <https://www.gnu.org/licenses/gpl-2.0.txt>
#
# Used to read files from a git repository without excessive forking.
use parent qw(Exporter);
use POSIX ();
use IO::Handle; # ->autoflush
-use Errno qw(EINTR);
+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);
+use PublicInbox::Spawn qw(popen_rd spawn);
use PublicInbox::Tmpfile;
+use IO::Poll qw(POLLIN);
use Carp qw(croak);
+use Digest::SHA ();
+use PublicInbox::DS qw(dwaitpid);
our @EXPORT_OK = qw(git_unquote git_quote);
our $PIPE_BUFSIZ = 65536; # Linux default
our $in_cleanup;
+our $RDTIMEO = 60_000; # milliseconds
-use constant MAX_INFLIGHT =>
- (($^O eq 'linux' ? 4096 : POSIX::_POSIX_PIPE_BUF()) * 3)
- /
+use constant MAX_INFLIGHT => (POSIX::PIPE_BUF * 3) /
65; # SHA-256 hex size + "\n" in preparation for git using non-SHA1
my %GIT_ESC = (
sub git_unquote ($) {
return $_[0] unless ($_[0] =~ /\A"(.*)"\z/);
$_[0] = $1;
- $_[0] =~ s/\\([\\"abfnrtv])/$GIT_ESC{$1}/g;
- $_[0] =~ s/\\([0-7]{1,3})/chr(oct($1))/ge;
+ $_[0] =~ s!\\([\\"abfnrtv]|[0-3][0-7]{2})!$GIT_ESC{$1}//chr(oct($1))!ge;
$_[0];
}
sub git_quote ($) {
if ($_[0] =~ s/([\\"\a\b\f\n\r\t\013]|[^[:print:]])/
- '\\'.($ESC_GIT{$1}||sprintf("%0o",ord($1)))/egs) {
+ '\\'.($ESC_GIT{$1}||sprintf("%03o",ord($1)))/egs) {
return qq{"$_[0]"};
}
$_[0];
sub new {
my ($class, $git_dir) = @_;
+ $git_dir =~ tr!/!/!s;
+ $git_dir =~ s!/*\z!!s;
# may contain {-tmp} field for File::Temp::Dir
bless { git_dir => $git_dir, alt_st => '', -git_path => {} }, $class
}
sub last_check_err {
my ($self) = @_;
my $fh = $self->{err_c} or return;
- sysseek($fh, 0, 0) or fail($self, "sysseek failed: $!");
+ sysseek($fh, 0, 0) or $self->fail("sysseek failed: $!");
defined(sysread($fh, my $buf, -s $fh)) or
- fail($self, "sysread failed: $!");
+ $self->fail("sysread failed: $!");
$buf;
}
if ($self->{$pid}) {
if (defined $err) { # "err_c"
my $fh = $self->{$err};
- sysseek($fh, 0, 0) or fail($self, "sysseek failed: $!");
- truncate($fh, 0) or fail($self, "truncate failed: $!");
+ sysseek($fh, 0, 0) or $self->fail("sysseek failed: $!");
+ truncate($fh, 0) or $self->fail("truncate failed: $!");
}
return;
}
- my ($out_r, $out_w);
- pipe($out_r, $out_w) or fail($self, "pipe failed: $!");
- my @cmd = (qw(git), "--git-dir=$self->{git_dir}",
+ pipe(my ($out_r, $out_w)) or $self->fail("pipe failed: $!");
+ my $rdr = { 0 => $out_r };
+ my $gd = $self->{git_dir};
+ if ($gd =~ s!/([^/]+/[^/]+)\z!/!) {
+ $rdr->{-C} = $gd;
+ $gd = $1;
+ }
+ my @cmd = (qw(git), "--git-dir=$gd",
qw(-c core.abbrev=40 cat-file), $batch);
- my $redir = { 0 => $out_r };
if ($err) {
my $id = "git.$self->{git_dir}$batch.err";
- my $fh = tmpfile($id) or fail($self, "tmpfile($id): $!");
+ my $fh = tmpfile($id) or $self->fail("tmpfile($id): $!");
$self->{$err} = $fh;
- $redir->{2} = $fh;
+ $rdr->{2} = $fh;
}
- my ($in_r, $p) = popen_rd(\@cmd, undef, $redir);
+ my ($in_r, $p) = popen_rd(\@cmd, undef, $rdr);
$self->{$pid} = $p;
+ $self->{"$pid.owner"} = $$;
$out_w->autoflush(1);
if ($^O eq 'linux') { # 1031: F_SETPIPE_SZ
fcntl($out_w, 1031, 4096);
$self->{$in} = $in_r;
}
+sub poll_in ($) { IO::Poll::_poll($RDTIMEO, fileno($_[0]), my $ev = POLLIN) }
+
sub my_read ($$$) {
my ($fh, $rbuf, $len) = @_;
my $left = $len - length($$rbuf);
$r = sysread($fh, $$rbuf, $PIPE_BUFSIZ, length($$rbuf));
if ($r) {
$left -= $r;
+ } elsif (defined($r)) { # EOF
+ return 0;
} else {
- next if (!defined($r) && $! == EINTR);
- return $r;
+ next if ($! == EAGAIN and poll_in($fh));
+ next if $! == EINTR; # may be set by sysread or poll_in
+ return; # unrecoverable error
}
}
\substr($$rbuf, 0, $len, '');
if ((my $n = index($$rbuf, "\n")) >= 0) {
return substr($$rbuf, 0, $n + 1, '');
}
- my $r = sysread($fh, $$rbuf, $PIPE_BUFSIZ, length($$rbuf));
- next if $r || (!defined($r) && $! == EINTR);
- return defined($r) ? '' : undef; # EOF or error
+ my $r = sysread($fh, $$rbuf, $PIPE_BUFSIZ, length($$rbuf))
+ and next;
+
+ # return whatever's left on EOF
+ return substr($$rbuf, 0, length($$rbuf)+1, '') if defined($r);
+
+ next if ($! == EAGAIN and poll_in($fh));
+ next if $! == EINTR; # may be set by sysread or poll_in
+ return; # unrecoverable error
}
}
for (my $i = 0; $i < @$inflight; $i += 3) {
$buf .= "$inflight->[$i]\n";
}
- print { $self->{out} } $buf or fail($self, "write error: $!");
+ print { $self->{out} } $buf or $self->fail("write error: $!");
unshift(@$inflight, \$req, $cb, $arg); # \$ref to indicate retried
cat_async_step($self, $inflight); # take one step
my ($self, $inflight) = @_;
die 'BUG: inflight empty or odd' if scalar(@$inflight) < 3;
my ($req, $cb, $arg) = splice(@$inflight, 0, 3);
- my $rbuf = delete($self->{cat_rbuf}) // \(my $new = '');
+ my $rbuf = delete($self->{rbuf}) // \(my $new = '');
my ($bref, $oid, $type, $size);
my $head = my_readline($self->{in}, $rbuf);
+ # ->fail may be called via Gcf2Client.pm
if ($head =~ /^([0-9a-f]{40,}) (\S+) ([0-9]+)$/) {
($oid, $type, $size) = ($1, $2, $3 + 0);
$bref = my_read($self->{in}, $rbuf, $size + 1) or
- fail($self, defined($bref) ? 'read EOF' : "read: $!");
- chop($$bref) eq "\n" or fail($self, 'LF missing after blob');
- } elsif ($head =~ / missing$/) {
+ $self->fail(defined($bref) ? 'read EOF' : "read: $!");
+ chop($$bref) eq "\n" or $self->fail('LF missing after blob');
+ } elsif ($head =~ s/ missing\n//s) {
+ $oid = $head;
# ref($req) indicates it's already been retried
# -gcf2 retries internally, so it never hits this path:
if (!ref($req) && !$in_cleanup && $self->alternates_changed) {
$req, $cb, $arg);
}
$type = 'missing';
- $oid = ref($req) ? $$req : $req;
+ $oid = ref($req) ? $$req : $req if $oid eq '';
} else {
- fail($self, "Unexpected result from async git cat-file: $head");
+ my $err = $! ? " ($!)" : '';
+ $self->fail("bad result from async cat-file: $head$err");
}
+ $self->{rbuf} = $rbuf if $$rbuf ne '';
eval { $cb->($bref, $oid, $type, $size, $arg) };
- $self->{cat_rbuf} = $rbuf if $$rbuf ne '';
warn "E: $oid: $@\n" if $@;
}
sub cat_async_wait ($) {
my ($self) = @_;
- my $inflight = delete $self->{inflight} or return;
+ my $inflight = $self->{inflight} or return;
while (scalar(@$inflight)) {
cat_async_step($self, $inflight);
}
}
sub _cat_file_cb {
- my ($bref, undef, undef, $size, $result) = @_;
- @$result = ($bref, $size);
+ my ($bref, $oid, $type, $size, $result) = @_;
+ @$result = ($bref, $oid, $type, $size);
}
sub cat_file {
- my ($self, $oid, $sizeref) = @_;
+ my ($self, $oid) = @_;
my $result = [];
cat_async($self, $oid, \&_cat_file_cb, $result);
cat_async_wait($self);
- $$sizeref = $result->[1] if $sizeref;
- $result->[0];
+ wantarray ? @$result : $result->[0];
}
sub check_async_step ($$) {
# https://public-inbox.org/git/20190118033845.s2vlrb3wd3m2jfzu@dcvr/T/
if ($hex eq 'dangling' || $hex eq 'notdir' || $hex eq 'loop') {
my $ret = my_read($self->{in_c}, $rbuf, $type + 1);
- fail($self, defined($ret) ? 'read EOF' : "read: $!") if !$ret;
+ $self->fail(defined($ret) ? 'read EOF' : "read: $!") if !$ret;
}
+ $self->{rbuf_c} = $rbuf if $$rbuf ne '';
eval { $cb->($hex, $type, $size, $arg, $self) };
warn "E: check($req) $@\n" if $@;
- $self->{rbuf_c} = $rbuf if $$rbuf ne '';
}
sub check_async_wait ($) {
my ($self) = @_;
- my $inflight_c = delete $self->{inflight_c} or return;
+ my $inflight_c = $self->{inflight_c} or return;
while (scalar(@$inflight_c)) {
check_async_step($self, $inflight_c);
}
sub check_async ($$$$) {
my ($self, $oid, $cb, $arg) = @_;
my $inflight_c = $self->{inflight_c} // check_async_begin($self);
- if (scalar(@$inflight_c) >= MAX_INFLIGHT) {
+ while (scalar(@$inflight_c) >= MAX_INFLIGHT) {
check_async_step($self, $inflight_c);
}
- print { $self->{out_c} } $oid, "\n" or fail($self, "write error: $!");
+ print { $self->{out_c} } $oid, "\n" or $self->fail("write error: $!");
push(@$inflight_c, $oid, $cb, $arg);
}
sub _destroy {
my ($self, $rbuf, $in, $out, $pid, $err) = @_;
- my $p = delete $self->{$pid} or return;
delete @$self{($rbuf, $in, $out)};
delete $self->{$err} if $err; # `err_c'
- # PublicInbox::DS may not be loaded
- eval { PublicInbox::DS::dwaitpid($p, undef, undef) };
- waitpid($p, 0) if $@; # wait synchronously if not in event loop
+ # GitAsyncCat::event_step may delete {pid}
+ my $p = delete $self->{$pid} or return;
+ dwaitpid($p) if $$ == $self->{"$pid.owner"};
}
-sub cat_async_abort ($) {
+sub async_abort ($) {
my ($self) = @_;
- my $inflight = delete $self->{inflight} or die 'BUG: not in async';
+ while (scalar(@{$self->{inflight_c} // []}) ||
+ scalar(@{$self->{inflight} // []})) {
+ for my $c ('', '_c') {
+ my $q = $self->{"inflight$c"};
+ while (@$q) {
+ my ($req, $cb, $arg) = splice(@$q, 0, 3);
+ $req =~ s/ .*//; # drop git_dir for Gcf2Client
+ eval { $cb->(undef, $req, undef, undef, $arg) };
+ warn "E: $req: $@ (in abort)\n" if $@;
+ }
+ delete $self->{"inflight$c"};
+ delete $self->{"rbuf$c"};
+ }
+ }
cleanup($self);
}
-sub fail {
+sub fail { # may be augmented in subclasses
my ($self, $msg) = @_;
- $self->{inflight} ? cat_async_abort($self) : cleanup($self);
- croak("git $self->{git_dir}: $msg");
+ async_abort($self);
+ croak(ref($self) . ' ' . ($self->{git_dir} // '') . ": $msg");
}
+# $git->popen(qw(show f00)); # or
+# $git->popen(qw(show f00), { GIT_CONFIG => ... }, { 2 => ... });
sub popen {
- my ($self, @cmd) = @_;
- @cmd = ('git', "--git-dir=$self->{git_dir}", @cmd);
- popen_rd(\@cmd);
+ my ($self, $cmd) = splice(@_, 0, 2);
+ $cmd = [ 'git', "--git-dir=$self->{git_dir}",
+ ref($cmd) ? @$cmd : ($cmd, grep { defined && !ref } @_) ];
+ popen_rd($cmd, grep { !defined || ref } @_); # env and opt
}
+# same args as popen above
sub qx {
- my ($self, @cmd) = @_;
- my $fh = $self->popen(@cmd);
- local $/ = "\n";
- return <$fh> if wantarray;
- local $/;
- <$fh>
+ my $fh = popen(@_);
+ if (wantarray) {
+ my @ret = <$fh>;
+ close $fh; # caller should check $?
+ @ret;
+ } else {
+ local $/;
+ my $ret = <$fh>;
+ close $fh; # caller should check $?
+ $ret;
+ }
+}
+
+sub date_parse {
+ my $self = shift;
+ map {
+ substr($_, length('--max-age='), -1)
+ } $self->qx('rev-parse', map { "--since=$_" } @_);
+}
+
+# check_async and cat_async may trigger the other, so ensure they're
+# both completely done by using this:
+sub async_wait_all ($) {
+ my ($self) = @_;
+ while (scalar(@{$self->{inflight_c} // []}) ||
+ scalar(@{$self->{inflight} // []})) {
+ check_async_wait($self);
+ cat_async_wait($self);
+ }
}
# returns true if there are pending "git cat-file" processes
sub cleanup {
- my ($self) = @_;
+ my ($self, $lazy) = @_;
+ return 1 if $lazy && (scalar(@{$self->{inflight_c} // []}) ||
+ scalar(@{$self->{inflight} // []}));
local $in_cleanup = 1;
delete $self->{async_cat};
- check_async_wait($self);
- cat_async_wait($self);
- _destroy($self, qw(cat_rbuf in out pid));
- _destroy($self, qw(chk_rbuf in_c out_c pid_c err_c));
- !!($self->{pid} || $self->{pid_c});
+ async_wait_all($self);
+ delete $self->{inflight};
+ delete $self->{inflight_c};
+ _destroy($self, qw(rbuf in out pid));
+ _destroy($self, qw(rbuf_c in_c out_c pid_c err_c));
+ undef;
}
# assuming a well-maintained repo, this should be a somewhat
my ($self) = @_;
my $ret = '???';
# don't show full FS path, basename should be OK:
- if ($self->{git_dir} =~ m!/([^/]+)(?:/\.git)?\z!) {
- $ret = "/path/to/$1";
+ if ($self->{git_dir} =~ m!/([^/]+)(?:/*\.git/*)?\z!) {
+ $ret = "$1.git";
}
wantarray ? ($ret) : $ret;
}
sub cat_async_begin {
my ($self) = @_;
cleanup($self) if $self->alternates_changed;
- batch_prepare($self);
+ $self->batch_prepare;
die 'BUG: already in async' if $self->{inflight};
$self->{inflight} = [];
}
sub cat_async ($$$;$) {
my ($self, $oid, $cb, $arg) = @_;
my $inflight = $self->{inflight} // cat_async_begin($self);
- if (scalar(@$inflight) >= MAX_INFLIGHT) {
+ while (scalar(@$inflight) >= MAX_INFLIGHT) {
cat_async_step($self, $inflight);
}
-
- print { $self->{out} } $oid, "\n" or fail($self, "write error: $!");
+ print { $self->{out} } $oid, "\n" or $self->fail("write error: $!");
push(@$inflight, $oid, $cb, $arg);
}
-# this is safe to call inside $cb, but not guaranteed to enqueue
-# returns true if successful, undef if not.
-sub async_prefetch {
- my ($self, $oid, $cb, $arg) = @_;
- if (defined($self->{async_cat}) && (my $inflight = $self->{inflight})) {
- # we could use MAX_INFLIGHT here w/o the halving,
- # but lets not allow one client to monopolize a git process
- if (scalar(@$inflight) < int(MAX_INFLIGHT/2)) {
- print { $self->{out} } $oid, "\n" or
- fail($self, "write error: $!");
- return push(@$inflight, $oid, $cb, $arg);
- }
+# returns the modified time of a git repo, same as the "modified" field
+# of a grokmirror manifest
+sub modified ($) {
+ # committerdate:unix is git 2.9.4+ (2017-05-05), so using raw instead
+ my $fh = popen($_[0], qw[for-each-ref --sort=-committerdate
+ --format=%(committerdate:raw) --count=1]);
+ (split(/ /, <$fh> // time))[0] + 0; # integerize for JSON
+}
+
+# for grokmirror, which doesn't read gitweb.description
+# templates/hooks--update.sample and git-multimail in git.git
+# only match "Unnamed repository", not the full contents of
+# templates/this--description in git.git
+sub manifest_entry {
+ my ($self, $epoch, $default_desc) = @_;
+ my $fh = $self->popen('show-ref');
+ my $dig = Digest::SHA->new(1);
+ while (read($fh, my $buf, 65536)) {
+ $dig->add($buf);
}
- undef;
-}
-
-sub extract_cmt_time {
- my ($bref, undef, undef, undef, $modified) = @_;
-
- if ($$bref =~ /^committer .*?> ([0-9]+) [\+\-]?[0-9]+/sm) {
- my $cmt_time = $1 + 0;
- $$modified = $cmt_time if $cmt_time > $$modified;
+ close $fh or return; # empty, uninitialized git repo
+ undef $fh; # for open, below
+ my $git_dir = $self->{git_dir};
+ my $ent = {
+ fingerprint => $dig->hexdigest,
+ reference => undef,
+ modified => modified($self),
+ };
+ chomp(my $owner = $self->qx('config', 'gitweb.owner'));
+ utf8::decode($owner);
+ $ent->{owner} = $owner eq '' ? undef : $owner;
+ my $desc = '';
+ if (open($fh, '<', "$git_dir/description")) {
+ local $/ = "\n";
+ chomp($desc = <$fh>);
+ utf8::decode($desc);
+ }
+ $desc = 'Unnamed repository' if $desc eq '';
+ if (defined $epoch && $desc =~ /\AUnnamed repository/) {
+ $desc = "$default_desc [epoch $epoch]";
+ }
+ $ent->{description} = $desc;
+ if (open($fh, '<', "$git_dir/objects/info/alternates")) {
+ # n.b.: GitPython doesn't seem to handle comments or C-quoted
+ # strings like native git does; and we don't for now, either.
+ local $/ = "\n";
+ chomp(my @alt = <$fh>);
+
+ # grokmirror only supports 1 alternate for "reference",
+ if (scalar(@alt) == 1) {
+ my $objdir = "$git_dir/objects";
+ my $ref = File::Spec->rel2abs($alt[0], $objdir);
+ $ref =~ s!/[^/]+/?\z!!; # basename
+ $ent->{reference} = $ref;
+ }
}
+ $ent;
}
-# returns the modified time of a git repo, same as the "modified" field
-# of a grokmirror manifest
-sub modified ($) {
+# returns true if there are pending cat-file processes
+sub cleanup_if_unlinked {
my ($self) = @_;
- my $modified = 0;
- my $fh = popen($self, qw(rev-parse --branches));
- local $/ = "\n";
- while (my $oid = <$fh>) {
- chomp $oid;
- cat_async($self, $oid, \&extract_cmt_time, \$modified);
+ return cleanup($self, 1) if $^O ne 'linux';
+ # Linux-specific /proc/$PID/maps access
+ # TODO: support this inside git.git
+ my $ret = 0;
+ for my $fld (qw(pid pid_c)) {
+ my $pid = $self->{$fld} // next;
+ open my $fh, '<', "/proc/$pid/maps" or return cleanup($self, 1);
+ while (<$fh>) {
+ # n.b. we do not restart for unlinked multi-pack-index
+ # since it's not too huge, and the startup cost may
+ # be higher.
+ /\.(?:idx|pack) \(deleted\)$/ and
+ return cleanup($self, 1);
+ }
+ ++$ret;
}
- cat_async_wait($self);
- $modified || time;
+ $ret;
}
1;