use File::Temp qw();
use Fcntl qw(SEEK_SET);
use PublicInbox::Git qw(git_unquote git_quote);
-use PublicInbox::Spawn qw(spawn popen_rd);
use PublicInbox::MsgIter qw(msg_iter msg_part_text);
+use PublicInbox::Qspawn;
+use PublicInbox::Tmpfile;
use URI::Escape qw(uri_escape_utf8);
+# POSIX requires _POSIX_ARG_MAX >= 4096, and xargs is required to
+# subtract 2048 bytes. We also don't factor in environment variable
+# headroom into this.
+use POSIX qw(sysconf _SC_ARG_MAX);
+my $ARG_SIZE_MAX = (sysconf(_SC_ARG_MAX) || 4096) - 2048;
+my $OID_MIN = 7;
+
+# By default, "git format-patch" generates filenames with a four-digit
+# prefix, so that means 9999 patch series are OK, right? :>
+# Maybe we can make this configurable, main concern is disk space overhead
+# for uncompressed patch fragments. Aside from space, public-inbox-httpd
+# is otherwise unaffected by having many patches, here, as it can share
+# work fairly. Other PSGI servers may have trouble, though.
+my $MAX_PATCH = 9999;
+
+# di = diff info / a hashref with information about a diff ($di):
+# {
+# oid_a => abbreviated pre-image oid,
+# oid_b => abbreviated post-image oid,
+# tmp => anonymous file handle with the diff,
+# hdr_lines => arrayref of various header lines for mode information
+# mode_a => original mode of oid_a (string, not integer),
+# ibx => PublicInbox::Inbox object containing the diff
+# smsg => PublicInbox::SearchMsg object containing diff
+# path_a => pre-image path
+# path_b => post-image path
+# n => numeric path of the patch (relative to worktree)
+# }
+
# don't bother if somebody sends us a patch with these path components,
# it's junk at best, an attack attempt at worse:
my %bad_component = map { $_ => 1 } ('', '.', '..');
-sub new {
- my ($class, $gits, $inboxes) = @_;
- bless {
- gits => $gits,
- inboxes => $inboxes,
- }, $class;
+sub dbg ($$) {
+ print { $_[0]->{out} } $_[1], "\n" or ERR($_[0], "print(dbg): $!");
}
-# look for existing blobs already in git repos
-sub solve_existing ($$$) {
- my ($self, $out, $want) = @_;
+sub ERR ($$) {
+ my ($self, $err) = @_;
+ print { $self->{out} } $err, "\n";
+ my $ucb = delete($self->{user_cb});
+ eval { $ucb->($err) } if $ucb;
+ die $err;
+}
+
+# look for existing objects already in git repos
+sub solve_existing ($$) {
+ my ($self, $want) = @_;
my $oid_b = $want->{oid_b};
+ my $have_hints = scalar keys %$want > 1;
my @ambiguous; # Array of [ git, $oids]
foreach my $git (@{$self->{gits}}) {
my ($oid_full, $type, $size) = $git->check($oid_b);
- if (defined($type) && $type eq 'blob') {
+ if (defined($type) && (!$have_hints || $type eq 'blob')) {
return [ $git, $oid_full, $type, int($size) ];
}
# TODO: do something with the ambiguous array?
# push @ambiguous, [ $git, @oids ];
- print $out "`$oid_b' ambiguous in ",
- join("\n", $git->pub_urls), "\n",
- join('', map { "$_ blob\n" } @oids), "\n";
+ dbg($self, "`$oid_b' ambiguous in " .
+ join("\n\t", $git->pub_urls($self->{psgi_env}))
+ . "\n" .
+ join('', map { "$_ blob\n" } @oids));
}
scalar(@ambiguous) ? \@ambiguous : undef;
}
-# returns a hashref with information about a diff ($di):
-# {
-# oid_a => abbreviated pre-image oid,
-# oid_b => abbreviated post-image oid,
-# tmp => anonymous file handle with the diff,
-# hdr_lines => arrayref of various header lines for mode information
-# mode_a => original mode of oid_a (string, not integer),
-# ibx => PublicInbox::Inbox object containing the diff
-# smsg => PublicInbox::SearchMsg object containing diff
-# path_a => pre-image path
-# path_b => post-image path
-# }
-sub extract_diff ($$$$) {
- my ($p, $re, $ibx, $smsg) = @_;
+sub extract_diff ($$$$$) {
+ my ($self, $p, $re, $ibx, $smsg) = @_;
my ($part) = @$p; # ignore $depth and @idx;
my $hdr_lines; # diff --git a/... b/...
my $tmp;
my ($s, undef) = msg_part_text($part, $ct);
defined $s or return;
my $di = {};
+
+ # Email::MIME::Encodings forces QP to be CRLF upon decoding,
+ # change it back to LF:
+ my $cte = $part->header('Content-Transfer-Encoding') || '';
+ if ($cte =~ /\bquoted-printable\b/i && $part->crlf eq "\n") {
+ $s =~ s/\r\n/\n/sg;
+ }
+
foreach my $l (split(/^/m, $s)) {
if ($l =~ $re) {
$di->{oid_a} = $1;
}
}
+
# start writing the diff out to a tempfile
- open($tmp, '+>', undef) or die "open(tmp): $!";
- $di->{tmp} = $tmp;
+ my $path = ++$self->{tot};
+ $di->{n} = $path;
+ open($tmp, '>', $self->{tmp}->dirname . "/$path") or
+ die "open(tmp): $!";
push @$hdr_lines, $l;
$di->{hdr_lines} = $hdr_lines;
- print $tmp @$hdr_lines, $l or die "print(tmp): $!";
+ utf8::encode($_) for @$hdr_lines;
+ print $tmp @$hdr_lines or die "print(tmp): $!";
# for debugging/diagnostics:
$di->{ibx} = $ibx;
$di->{smsg} = $smsg;
- } elsif ($l =~ m!\Adiff --git ("?a/.+) ("?b/.+)$!) {
- return $di if $tmp; # got our blob, done!
+ } elsif ($l =~ m!\Adiff --git ("?[^/]+/.+) ("?[^/]+/.+)$!) {
+ last if $tmp; # got our blob, done!
my ($path_a, $path_b) = ($1, $2);
+ # diff header lines won't have \r because git
+ # will quote them, but Email::MIME gives CRLF
+ # for quoted-printable:
+ $path_b =~ tr/\r//d;
+
# don't care for leading 'a/' and 'b/'
my (undef, @a) = split(m{/}, git_unquote($path_a));
my (undef, @b) = split(m{/}, git_unquote($path_b));
$di->{path_b} = join('/', @b);
$hdr_lines = [ $l ];
} elsif ($tmp) {
+ utf8::encode($l);
print $tmp $l or die "print(tmp): $!";
} elsif ($hdr_lines) {
push @$hdr_lines, $l;
}
}
}
- $tmp ? $di : undef;
+ return undef unless $tmp;
+ close $tmp or die "close(tmp): $!";
+ $di;
}
sub path_searchable ($) { defined($_[0]) && $_[0] =~ m!\A[\w/\. \-]+\z! }
-sub find_extract_diff ($$$) {
+# ".." appears in path names, which confuses Xapian into treating
+# it as a range query. So we split on ".." since Xapian breaks
+# on punctuation anyways:
+sub filename_query ($) {
+ join('', map { qq( dfn:"$_") } split(/\.\./, $_[0]));
+}
+
+sub find_extract_diffs ($$$) {
my ($self, $ibx, $want) = @_;
my $srch = $ibx->search or return;
my $path_b = $want->{path_b};
if (path_searchable($path_b)) {
- $q .= qq{ dfn:"$path_b"};
+ $q .= filename_query($path_b);
my $path_a = $want->{path_a};
if (path_searchable($path_a) && $path_a ne $path_b) {
- $q .= qq{ dfn:"$path_a"};
+ $q .= filename_query($path_a);
}
}
my $msgs = $srch->query($q, { relevance => 1 });
- my $re = qr/\Aindex ($pre[a-f0-9]*)\.\.($post[a-f0-9]*)(?: (\d+))?/;
+ my $re = qr/\Aindex ($pre[a-f0-9]*)\.\.($post[a-f0-9]*)(?: ([0-9]+))?/;
- my $di;
+ my @di;
foreach my $smsg (@$msgs) {
$ibx->smsg_mime($smsg) or next;
msg_iter(delete($smsg->{mime}), sub {
- $di ||= extract_diff($_[0], $re, $ibx, $smsg);
+ my $di = extract_diff($self, $_[0], $re, $ibx, $smsg);
+ push @di, $di if defined($di);
});
- return $di if $di;
}
+ @di ? \@di : undef;
+}
+
+sub prepare_index ($) {
+ my ($self) = @_;
+ my $patches = $self->{patches};
+ $self->{nr} = 0;
+
+ my $di = $patches->[0] or die 'no patches';
+ my $oid_a = $di->{oid_a} or die '{oid_a} unset';
+ my $existing = $self->{found}->{$oid_a};
+
+ # no index creation for added files
+ $oid_a =~ /\A0+\z/ and return next_step($self);
+
+ die "BUG: $oid_a not not found" unless $existing;
+
+ my $oid_full = $existing->[1];
+ my $path_a = $di->{path_a} or die "BUG: path_a missing for $oid_full";
+ my $mode_a = $di->{mode_a} || extract_old_mode($di);
+
+ my $in = tmpfile("update-index.$oid_full") or die "tmpfile: $!";
+ print $in "$mode_a $oid_full\t$path_a\0" or die "print: $!";
+ $in->flush or die "flush: $!";
+ sysseek($in, 0, 0) or die "seek: $!";
+
+ dbg($self, 'preparing index');
+ my $rdr = { 0 => fileno($in), -hold => $in };
+ my $cmd = [ qw(git update-index -z --index-info) ];
+ my $qsp = PublicInbox::Qspawn->new($cmd, $self->{git_env}, $rdr);
+ $qsp->psgi_qx($self->{psgi_env}, undef, sub {
+ my ($bref) = @_;
+ if (my $err = $qsp->{err}) {
+ ERR($self, "git update-index error: $err");
+ }
+ dbg($self, "index prepared:\n" .
+ "$mode_a $oid_full\t" . git_quote($path_a));
+ next_step($self); # onto do_git_apply
+ });
}
# pure Perl "git init"
-sub do_git_init_wt ($) {
+sub do_git_init ($) {
my ($self) = @_;
- my $wt = File::Temp->newdir('solver.wt-XXXXXXXX', TMPDIR => 1);
- my $dir = $wt->dirname;
+ my $dir = $self->{tmp}->dirname;
+ my $git_dir = "$dir/git";
foreach ('', qw(objects refs objects/info refs/heads)) {
- mkdir("$dir/.git/$_") or die "mkdir $_: $!";
+ mkdir("$git_dir/$_") or die "mkdir $_: $!";
}
- open my $fh, '>', "$dir/.git/config" or die "open .git/config: $!";
- print $fh <<'EOF' or die "print .git/config $!";
+ open my $fh, '>', "$git_dir/config" or die "open git/config: $!";
+ print $fh <<'EOF' or die "print git/config $!";
[core]
repositoryFormatVersion = 0
filemode = true
fsyncObjectfiles = false
logAllRefUpdates = false
EOF
- close $fh or die "close .git/config: $!";
+ close $fh or die "close git/config: $!";
- open $fh, '>', "$dir/.git/HEAD" or die "open .git/HEAD: $!";
- print $fh "ref: refs/heads/master\n" or die "print .git/HEAD: $!";
- close $fh or die "close .git/HEAD: $!";
+ open $fh, '>', "$git_dir/HEAD" or die "open git/HEAD: $!";
+ print $fh "ref: refs/heads/master\n" or die "print git/HEAD: $!";
+ close $fh or die "close git/HEAD: $!";
- my $f = '.git/objects/info/alternates';
- open $fh, '>', "$dir/$f" or die "open: $f: $!";
- print($fh (map { "$_->{git_dir}/objects\n" } @{$self->{gits}})) or
- die "print $f: $!";
+ my $f = 'objects/info/alternates';
+ open $fh, '>', "$git_dir/$f" or die "open: $f: $!";
+ foreach my $git (@{$self->{gits}}) {
+ print $fh $git->git_path('objects'),"\n" or die "print $f: $!";
+ }
close $fh or die "close: $f: $!";
- $wt;
+ my $tmp_git = $self->{tmp_git} = PublicInbox::Git->new($git_dir);
+ $tmp_git->{-tmp} = $self->{tmp};
+ $self->{git_env} = {
+ GIT_DIR => $git_dir,
+ GIT_INDEX_FILE => "$git_dir/index",
+ };
+ prepare_index($self);
}
sub extract_old_mode ($) {
my ($di) = @_;
- if (grep(/\Aold mode (100644|100755|120000)$/, @{$di->{hdr_lines}})) {
+ if (join('', @{$di->{hdr_lines}}) =~
+ /^old mode (100644|100755|120000)\b/) {
return $1;
}
'100644';
}
-sub reap ($$) {
- my ($pid, $msg) = @_;
- waitpid($pid, 0) == $pid or die "waitpid($msg): $!";
- $? == 0 or die "$msg failed: $?";
-}
-
-sub prepare_index ($$$$) {
- my ($out, $wt_dir, $existing, $di) = @_;
- my $oid_full = $existing->[1];
- my ($r, $w);
- my $path_a = $di->{path_a} or die "BUG: path_a missing for $oid_full";
- my $mode_a = $di->{mode_a} || extract_old_mode($di);
-
- # unlike git-apply(1), this only gets called once in a patch
- # series and happens too quickly to be worth making async:
- pipe($r, $w) or die "pipe: $!";
- my $rdr = { 0 => fileno($r) };
- my $pid = spawn([qw(git -C), $wt_dir,
- qw(update-index -z --index-info)], undef, $rdr);
- close $r or die "close pipe(r): $!";
- print $w "$mode_a $oid_full\t$path_a\0" or die "print update-index: $!";
-
- close $w or die "close update-index: $!";
- reap($pid, 'update-index -z --index-info');
+sub do_finish ($$) {
+ my ($self, $user_cb) = @_;
+ my $found = $self->{found};
+ my $oid_want = $self->{oid_want};
+ if (my $exists = $found->{$oid_want}) {
+ return $user_cb->($exists);
+ }
- print $out "index prepared:\n",
- "$mode_a $oid_full\t", git_quote($path_a), "\n";
+ # let git disambiguate if oid_want was too short,
+ # but long enough to be unambiguous:
+ my $tmp_git = $self->{tmp_git};
+ if (my @res = $tmp_git->check($oid_want)) {
+ return $user_cb->($found->{$res[0]});
+ }
+ if (my $err = $tmp_git->last_check_err) {
+ dbg($self, $err);
+ }
+ $user_cb->(undef);
}
-sub do_apply_begin ($$$) {
- my ($out, $wt_dir, $di) = @_;
+sub do_step ($) {
+ my ($self) = @_;
+ eval {
+ # step 1: resolve blobs to patches in the todo queue
+ if (my $want = pop @{$self->{todo}}) {
+ # this populates {patches} and {todo}
+ resolve_patch($self, $want);
+
+ # step 2: then we instantiate a working tree once
+ # the todo queue is finally empty:
+ } elsif (!defined($self->{tmp_git})) {
+ do_git_init($self);
+
+ # step 3: apply each patch in the stack
+ } elsif (scalar @{$self->{patches}}) {
+ do_git_apply($self);
+
+ # step 4: execute the user-supplied callback with
+ # our result: (which may be undef)
+ # Other steps may call user_cb to terminate prematurely
+ # on error
+ } elsif (my $user_cb = delete($self->{user_cb})) {
+ do_finish($self, $user_cb);
+ } else {
+ die 'about to call user_cb twice'; # Oops :x
+ }
+ }; # eval
+ my $err = $@;
+ if ($err) {
+ $err =~ s/^\s*Exception:\s*//; # bad word to show users :P
+ dbg($self, "E: $err");
+ my $ucb = delete($self->{user_cb});
+ eval { $ucb->($err) } if $ucb;
+ }
+}
- my $tmp = delete $di->{tmp} or die "BUG: no tmp ", di_url($di);
- $tmp->flush or die "tmp->flush failed: $!";
- $out->flush or die "err->flush failed: $!";
- sysseek($tmp, 0, SEEK_SET) or die "sysseek(tmp) failed: $!";
+sub step_cb ($) {
+ my ($self) = @_;
+ sub { do_step($self) };
+}
- defined(my $err_fd = fileno($out)) or die "fileno(out): $!";
- my $rdr = { 0 => fileno($tmp), 1 => $err_fd, 2 => $err_fd };
- my $cmd = [ qw(git -C), $wt_dir,
- qw(apply --cached --whitespace=warn --verbose) ];
- spawn($cmd, undef, $rdr);
+sub next_step ($) {
+ my ($self) = @_;
+ # if outside of public-inbox-httpd, caller is expected to be
+ # looping step_cb, anyways
+ my $async = $self->{psgi_env}->{'pi-httpd.async'} or return;
+ # PublicInbox::HTTPD::Async->new
+ $async->(undef, step_cb($self));
}
-sub do_apply_continue ($$) {
- my ($wt_dir, $apply_pid) = @_;
- reap($apply_pid, 'apply');
- popen_rd([qw(git -C), $wt_dir, qw(ls-files -s -z)]);
+sub mark_found ($$$) {
+ my ($self, $oid, $found_info) = @_;
+ my $found = $self->{found};
+ $found->{$oid} = $found_info;
+ my $oid_cur = $found_info->[1];
+ while ($oid_cur ne $oid && length($oid_cur) > $OID_MIN) {
+ $found->{$oid_cur} = $found_info;
+ chop($oid_cur);
+ }
}
-sub do_apply_end ($$$$) {
- my ($out, $wt_git, $rd, $di) = @_;
+sub parse_ls_files ($$$$) {
+ my ($self, $qsp, $bref, $di) = @_;
+ if (my $err = $qsp->{err}) {
+ die "git ls-files error: $err";
+ }
- local $/ = "\0";
- defined(my $line = <$rd>) or die "failed to read ls-files: $!";
- chomp $line or die "no trailing \\0 in [$line] from ls-files";
+ my ($line, @extra) = split(/\0/, $$bref);
+ scalar(@extra) and die "BUG: extra files in index: <",
+ join('> <', @extra), ">";
my ($info, $file) = split(/\t/, $line, 2);
my ($mode_b, $oid_b_full, $stage) = split(/ /, $info);
+ if ($file ne $di->{path_b}) {
+ die
+"BUG: index mismatch: file=$file != path_b=$di->{path_b}";
+ }
- defined($line = <$rd>) and die "extra files in index: $line";
- close $rd or die "close ls-files: $?";
+ my $tmp_git = $self->{tmp_git} or die 'no git working tree';
+ my (undef, undef, $size) = $tmp_git->check($oid_b_full);
+ defined($size) or die "check $oid_b_full failed";
- $file eq $di->{path_b} or
- die "index mismatch: file=$file != path_b=$di->{path_b}";
+ dbg($self, "index at:\n$mode_b $oid_b_full\t$file");
+ my $created = [ $tmp_git, $oid_b_full, 'blob', $size, $di ];
+ mark_found($self, $di->{oid_b}, $created);
+ next_step($self); # onto the next patch
+}
- my (undef, undef, $size) = $wt_git->check($oid_b_full);
+sub start_ls_files ($$) {
+ my ($self, $di) = @_;
+ my $cmd = [qw(git ls-files -s -z)];
+ my $qsp = PublicInbox::Qspawn->new($cmd, $self->{git_env});
+ $qsp->psgi_qx($self->{psgi_env}, undef, sub {
+ my ($bref) = @_;
+ eval { parse_ls_files($self, $qsp, $bref, $di) };
+ ERR($self, $@) if $@;
+ });
+}
+
+sub oids_same_ish ($$) {
+ (index($_[0], $_[1]) == 0) || (index($_[1], $_[0]) == 0);
+}
- defined($size) or die "failed to read_size from $oid_b_full";
+sub skip_identical ($$$) {
+ my ($self, $patches, $cur_oid_b) = @_;
+ while (my $nxt = $patches->[0]) {
+ if (oids_same_ish($cur_oid_b, $nxt->{oid_b})) {
+ dbg($self, 'skipping '.di_url($self, $nxt).
+ " for $cur_oid_b");
+ shift @$patches;
+ } else {
+ return;
+ }
+ }
+}
- print $out "$mode_b $oid_b_full\t$file\n";
- [ $wt_git, $oid_b_full, 'blob', $size, $di ];
+sub do_git_apply ($) {
+ my ($self) = @_;
+ my $dn = $self->{tmp}->dirname;
+ my $patches = $self->{patches};
+
+ # we need --ignore-whitespace because some patches are CRLF
+ my @cmd = (qw(git -C), $dn, qw(apply --cached --ignore-whitespace
+ --whitespace=warn --verbose));
+ my $len = length(join(' ', @cmd));
+ my $total = $self->{tot};
+ my $di; # keep track of the last one for "git ls-files"
+ my $prv_oid_b;
+
+ do {
+ my $i = ++$self->{nr};
+ $di = shift @$patches;
+ dbg($self, "\napplying [$i/$total] " . di_url($self, $di) .
+ "\n" . join('', @{$di->{hdr_lines}}));
+ my $path = $di->{n};
+ $len += length($path) + 1;
+ push @cmd, $path;
+ $prv_oid_b = $di->{oid_b};
+ } while (@$patches && $len < $ARG_SIZE_MAX &&
+ !oids_same_ish($patches->[0]->{oid_b}, $prv_oid_b));
+
+ my $rdr = { 2 => 1 };
+ my $qsp = PublicInbox::Qspawn->new(\@cmd, $self->{git_env}, $rdr);
+ $qsp->psgi_qx($self->{psgi_env}, undef, sub {
+ my ($bref) = @_;
+ dbg($self, $$bref);
+ if (my $err = $qsp->{err}) {
+ my $msg = "git apply error: $err";
+ my $nxt = $patches->[0];
+ if ($nxt && oids_same_ish($nxt->{oid_b}, $prv_oid_b)) {
+ dbg($self, $msg);
+ dbg($self, 'trying '.di_url($self, $nxt));
+ } else {
+ ERR($self, $msg);
+ }
+ } else {
+ skip_identical($self, $patches, $di->{oid_b});
+ }
+ eval { start_ls_files($self, $di) };
+ ERR($self, $@) if $@;
+ });
}
-sub di_url ($) {
- my ($di) = @_;
- # note: we don't pass the PSGI env here, different inboxes
- # can have different HTTP_HOST on the same instance.
- my $url = $di->{ibx}->base_url;
+sub di_url ($$) {
+ my ($self, $di) = @_;
+ # note: we don't pass the PSGI env unconditionally, here,
+ # different inboxes can have different HTTP_HOST on the same instance.
+ my $ibx = $di->{ibx};
+ my $env = $self->{psgi_env} if $ibx eq $self->{inboxes}->[0];
+ my $url = $ibx->base_url($env);
my $mid = $di->{smsg}->{mid};
defined($url) ? "$url$mid/" : "<$mid>";
}
-# reconstruct the oid_b blob using patches we found:
-sub apply_patches_cb ($$$$$) {
- my ($self, $out, $found, $patches, $oid_b) = @_;
+sub resolve_patch ($$) {
+ my ($self, $want) = @_;
- my $tot = scalar(@$patches) or return sub {
- print $out "no patch(es) for $oid_b\n";
- undef;
- };
+ if (scalar(@{$self->{patches}}) > $MAX_PATCH) {
+ die "Aborting, too many steps to $self->{oid_want}";
+ }
- my $wt = do_git_init_wt($self);
- my $wt_dir = $wt->dirname;
- my $wt_git = PublicInbox::Git->new("$wt_dir/.git");
- $wt_git->{-wt} = $wt;
-
- my $cur = 0;
- my ($apply_pid, $rd, $di);
-
- # returns an empty string if in progress, undef if not found,
- # or the final [ ::Git, oid_full, type, size, $di ] arrayref
- # if found
- sub {
- if ($rd) {
- $found->{$di->{oid_b}} =
- do_apply_end($out, $wt_git, $rd, $di);
- $rd = undef;
- # continue to shift @$patches
- } elsif ($apply_pid) {
- $rd = do_apply_continue($wt_dir, $apply_pid);
- $apply_pid = undef;
- return ''; # $rd => do_apply_ned
+ # see if we can find the blob in an existing git repo:
+ my $cur_want = $want->{oid_b};
+ if ($self->{seen_oid}->{$cur_want}++) {
+ die "Loop detected solving $cur_want\n";
+ }
+ if (my $existing = solve_existing($self, $want)) {
+ my ($found_git, undef, $type, undef) = @$existing;
+ dbg($self, "found $cur_want in " .
+ join("\n", $found_git->pub_urls($self->{psgi_env})));
+
+ if ($cur_want eq $self->{oid_want} || $type ne 'blob') {
+ eval { delete($self->{user_cb})->($existing) };
+ die "E: $@" if $@;
+ return;
}
+ mark_found($self, $cur_want, $existing);
+ return next_step($self); # onto patch application
+ }
- # may return undef here
- $di = shift @$patches or return $found->{$oid_b};
+ # scan through inboxes to look for emails which results in
+ # the oid we want:
+ foreach my $ibx (@{$self->{inboxes}}) {
+ my $diffs = find_extract_diffs($self, $ibx, $want) or next;
- my $i = ++$cur;
- my $oid_a = $di->{oid_a};
- my $existing = $found->{$oid_a};
- my $empty_oid = $oid_a =~ /\A0+\z/;
+ unshift @{$self->{patches}}, @$diffs;
+ dbg($self, "found $cur_want in ".
+ join("\n\t", map { di_url($self, $_) } @$diffs));
- if ($empty_oid && $i != 1) {
- die "empty oid at [$i/$tot] ", di_url($di);
- }
- if (!$existing && !$empty_oid) {
- die "missing $oid_a at [$i/$tot] ", di_url($di);
- }
+ # good, we can find a path to the oid we $want, now
+ # lets see if we need to apply more patches:
+ my $di = $diffs->[0];
+ my $src = $di->{oid_a};
- # prepare the worktree for patch application:
- if ($i == 1 && $existing) {
- prepare_index($out, $wt_dir, $existing, $di);
+ unless ($src =~ /\A0+\z/) {
+ # we have to solve it using another oid, fine:
+ my $job = { oid_b => $src, path_b => $di->{path_a} };
+ push @{$self->{todo}}, $job;
}
-
- print $out "\napplying [$i/$tot] ", di_url($di), "\n",
- join('', @{$di->{hdr_lines}}), "\n"
- or die "print \$out failed: $!";
-
- # begin the patch application patch!
- $apply_pid = do_apply_begin($out, $wt_dir, $di);
- # next call to this callback will call do_apply_continue
- '';
+ return next_step($self); # onto the next todo item
+ }
+ if (length($cur_want) > $OID_MIN) {
+ chop($cur_want);
+ dbg($self, "retrying $want->{oid_b} as $cur_want");
+ $want->{oid_b} = $cur_want;
+ push @{$self->{todo}}, $want;
+ return next_step($self); # retry with shorter abbrev
}
-}
-
-# recreate $oid_b
-# Returns an array ref: [ ::Git object, oid_full, type, size, di ]
-# or undef if nothing was found.
-sub solve ($$$$) {
- my ($self, $out, $oid_b, $hints) = @_;
-
- # should we even get here? Probably not, but somebody
- # could be manually typing URLs:
- return if $oid_b =~ /\A0+\z/;
-
- my $req = { %$hints, oid_b => $oid_b };
- my @todo = ($req);
- my $found = {}; # { abbrev => [ ::Git, oid_full, type, size, $di ] }
- my $patches = []; # [ array of $di hashes ]
- my $max = $self->{max_patches} || 200;
- my $apply_cb;
- my $cb = sub {
- my $want = pop @todo;
- unless ($want) {
- $apply_cb ||= apply_patches_cb($self, $out, $found,
- $patches, $oid_b);
- return $apply_cb->();
- }
-
- if (scalar(@$patches) > $max) {
- print $out "Aborting, too many steps to $oid_b\n";
- return;
- }
- # see if we can find the blob in an existing git repo:
- my $want_oid = $want->{oid_b};
- if (my $existing = solve_existing($self, $out, $want)) {
- print $out "found $want_oid in ",
- join("\n", $existing->[0]->pub_urls), "\n";
-
- return $existing if $want_oid eq $oid_b; # DONE!
- $found->{$want_oid} = $existing;
- return ''; # ok, one blob resolved, more to go?
- }
- # scan through inboxes to look for emails which results in
- # the oid we want:
- my $di;
- foreach my $ibx (@{$self->{inboxes}}) {
- $di = find_extract_diff($self, $ibx, $want) or next;
+ dbg($self, "could not find $cur_want");
+ eval { delete($self->{user_cb})->(undef) }; # not found! :<
+ die "E: $@" if $@;
+}
- unshift @$patches, $di;
- print $out "found $want_oid in ",di_url($di),"\n";
+# this API is designed to avoid creating self-referential structures;
+# so user_cb never references the SolverGit object
+sub new {
+ my ($class, $ibx, $user_cb) = @_;
- # good, we can find a path to the oid we $want, now
- # lets see if we need to apply more patches:
- my $src = $di->{oid_a};
+ bless {
+ gits => $ibx->{-repo_objs},
+ user_cb => $user_cb,
- last if $src =~ /\A0+\z/;
+ # TODO: config option for searching related inboxes
+ inboxes => [ $ibx ],
+ }, $class;
+}
- # we have to solve it using another oid, fine:
- my $job = { oid_b => $src, path_b => $di->{path_a} };
- push @todo, $job;
- last; # onto the next @todo item
- }
- unless ($di) {
- print $out "$want_oid could not be found\n";
- return;
- }
- ''; # continue onto next @todo item;
- };
+# recreate $oid_want using $hints
+# hints keys: path_a, path_b, oid_a
+# Calls {user_cb} with: [ ::Git object, oid_full, type, size, di (diff_info) ]
+# with found object, or undef if nothing was found
+# Calls {user_cb} with a string error on fatal errors
+sub solve ($$$$$) {
+ my ($self, $env, $out, $oid_want, $hints) = @_;
- while (1) {
- my $ret = $cb->();
- return $ret if (ref($ret) || !defined($ret));
- # $ret == ''; so continue looping here
+ # should we even get here? Probably not, but somebody
+ # could be manually typing URLs:
+ return (delete $self->{user_cb})->(undef) if $oid_want =~ /\A0+\z/;
+
+ $self->{oid_want} = $oid_want;
+ $self->{out} = $out;
+ $self->{seen_oid} = {};
+ $self->{tot} = 0;
+ $self->{psgi_env} = $env;
+ $self->{todo} = [ { %$hints, oid_b => $oid_want } ];
+ $self->{patches} = []; # [ $di, $di, ... ]
+ $self->{found} = {}; # { abbr => [ ::Git, oid, type, size, $di ] }
+ $self->{tmp} = File::Temp->newdir("solver.$oid_want-XXXXXXXX", TMPDIR => 1);
+
+ dbg($self, "solving $oid_want ...");
+ my $step_cb = step_cb($self);
+ if (my $async = $env->{'pi-httpd.async'}) {
+ # PublicInbox::HTTPD::Async->new
+ $async->(undef, $step_cb);
+ } else {
+ $step_cb->() while $self->{user_cb};
}
}