X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FSolverGit.pm;h=eea59b6ddbda047496fe4917517fe4b3c6c53bdb;hb=48d4fdb2acbf8cf05dcb357df53b0d658f88fe39;hp=e30720277d92e5cd2ea097a14f8c40686851e84c;hpb=8a7653c1e4b1305b95db7e65c7ad00d2a50bbce1;p=public-inbox.git diff --git a/lib/PublicInbox/SolverGit.pm b/lib/PublicInbox/SolverGit.pm index e3072027..eea59b6d 100644 --- a/lib/PublicInbox/SolverGit.pm +++ b/lib/PublicInbox/SolverGit.pm @@ -10,14 +10,29 @@ package PublicInbox::SolverGit; use strict; use warnings; -use File::Temp qw(); +use File::Temp 0.19 (); 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, @@ -29,6 +44,7 @@ use URI::Escape qw(uri_escape_utf8); # 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, @@ -47,14 +63,15 @@ sub ERR ($$) { die $err; } -# look for existing blobs already in git repos +# look for existing objects already in git repos sub solve_existing ($$) { 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) ]; } @@ -69,14 +86,15 @@ sub solve_existing ($$) { # push @ambiguous, [ $git, @oids ]; dbg($self, "`$oid_b' ambiguous in " . - join("\n\t", $git->pub_urls) . "\n" . + join("\n\t", $git->pub_urls($self->{psgi_env})) + . "\n" . join('', map { "$_ blob\n" } @oids)); } scalar(@ambiguous) ? \@ambiguous : undef; } -sub extract_diff ($$$$) { - my ($p, $re, $ibx, $smsg) = @_; +sub extract_diff ($$$$$) { + my ($self, $p, $re, $ibx, $smsg) = @_; my ($part) = @$p; # ignore $depth and @idx; my $hdr_lines; # diff --git a/... b/... my $tmp; @@ -103,19 +121,23 @@ sub extract_diff ($$$$) { } } + # start writing the diff out to a tempfile - open($tmp, '+>', undef) or die "open(tmp): $!"; - $di->{tmp} = $tmp; + my $path = ++$self->{tot}; + $di->{n} = $path; + open($tmp, '>', $self->{tmp}->dirname . "/$path") or + die "open(tmp): $!"; push @$hdr_lines, $l; $di->{hdr_lines} = $hdr_lines; + utf8::encode($_) for @$hdr_lines; print $tmp @$hdr_lines or die "print(tmp): $!"; # for debugging/diagnostics: $di->{ibx} = $ibx; $di->{smsg} = $smsg; - } elsif ($l =~ m!\Adiff --git ("?a/.+) ("?b/.+)$!) { - return $di if $tmp; # got our blob, done! + } elsif ($l =~ m!\Adiff --git ("?[^/]+/.+) ("?[^/]+/.+)$!) { + last if $tmp; # got our blob, done! my ($path_a, $path_b) = ($1, $2); @@ -137,6 +159,7 @@ sub extract_diff ($$$$) { $di->{path_b} = join('/', @b); $hdr_lines = [ $l ]; } elsif ($tmp) { + utf8::encode($l); print $tmp $l or die "print(tmp): $!"; } elsif ($hdr_lines) { push @$hdr_lines, $l; @@ -145,12 +168,21 @@ sub extract_diff ($$$$) { } } } - $tmp ? $di : undef; + return undef unless $tmp; + close $tmp or die "close(tmp): $!"; + $di; } sub path_searchable ($) { defined($_[0]) && $_[0] =~ m!\A[\w/\. \-]+\z! } -sub find_extract_diff ($$$) { +# ".." 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 ($$$) { my ($self, $ibx, $want) = @_; my $srch = $ibx->search or return; @@ -167,32 +199,32 @@ sub find_extract_diff ($$$) { 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; - $self->{tot} = scalar @$patches; my $di = $patches->[0] or die 'no patches'; my $oid_a = $di->{oid_a} or die '{oid_a} unset'; @@ -207,13 +239,13 @@ sub prepare_index ($) { 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); - open my $in, '+>', undef or die "open: $!"; + 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) }; + 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 { @@ -253,8 +285,9 @@ EOF my $f = 'objects/info/alternates'; open $fh, '>', "$git_dir/$f" or die "open: $f: $!"; - print($fh (map { "$_->{git_dir}/objects\n" } @{$self->{gits}})) or - die "print $f: $!"; + foreach my $git (@{$self->{gits}}) { + print $fh $git->git_path('objects'),"\n" or die "print $f: $!"; + } close $fh or die "close: $f: $!"; my $tmp_git = $self->{tmp_git} = PublicInbox::Git->new($git_dir); $tmp_git->{-tmp} = $self->{tmp}; @@ -267,12 +300,33 @@ EOF sub extract_old_mode ($) { 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 do_finish ($$) { + my ($self, $user_cb) = @_; + my $found = $self->{found}; + my $oid_want = $self->{oid_want}; + if (my $exists = $found->{$oid_want}) { + return $user_cb->($exists); + } + + # let git disambiguate if oid_want was too short, + # but long enough to be unambiguous: + my $tmp_git = $self->{tmp_git}; + if (my @res = $tmp_git->check($oid_want)) { + return $user_cb->($found->{$res[0]}); + } + if (my $err = $tmp_git->last_check_err) { + dbg($self, $err); + } + $user_cb->(undef); +} + sub do_step ($) { my ($self) = @_; eval { @@ -294,8 +348,8 @@ sub do_step ($) { # our result: (which may be undef) # Other steps may call user_cb to terminate prematurely # on error - } elsif (my $ucb = delete($self->{user_cb})) { - $ucb->($self->{found}->{$self->{oid_want}}); + } elsif (my $user_cb = delete($self->{user_cb})) { + do_finish($self, $user_cb); } else { die 'about to call user_cb twice'; # Oops :x } @@ -325,7 +379,13 @@ sub next_step ($) { sub mark_found ($$$) { my ($self, $oid, $found_info) = @_; - $self->{found}->{$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 parse_ls_files ($$$$) { @@ -366,29 +426,64 @@ sub start_ls_files ($$) { }); } -sub do_git_apply ($) { - my ($self) = @_; +sub oids_same_ish ($$) { + (index($_[0], $_[1]) == 0) || (index($_[1], $_[0]) == 0); +} - my $di = shift @{$self->{patches}} or die 'empty {patches}'; - my $tmp = delete $di->{tmp} or die 'no tmp ', di_url($self, $di); - $tmp->flush or die "tmp->flush failed: $!"; - sysseek($tmp, 0, SEEK_SET) or die "sysseek(tmp) failed: $!"; +sub skip_identical ($$$) { + 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; + } + } +} - my $i = ++$self->{nr}; - dbg($self, "\napplying [$i/$self->{tot}] " . di_url($self, $di) . - "\n" . join('', @{$di->{hdr_lines}})); +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 apply --cached --ignore-whitespace - --whitespace=warn --verbose) ]; - my $rdr = { 0 => fileno($tmp), 2 => 1 }; - my $qsp = PublicInbox::Qspawn->new($cmd, $self->{git_env}, $rdr); + my @cmd = (qw(git -C), $dn, qw(apply --cached --ignore-whitespace + --unidiff-zero --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) = @_; - close $tmp; dbg($self, $$bref); if (my $err = $qsp->{err}) { - ERR($self, "git apply error: $err"); + my $msg = "git apply error: $err"; + my $nxt = $patches->[0]; + if ($nxt && oids_same_ish($nxt->{oid_b}, $prv_oid_b)) { + dbg($self, $msg); + dbg($self, 'trying '.di_url($self, $nxt)); + } else { + ERR($self, $msg); + } + } else { + skip_identical($self, $patches, $di->{oid_b}); } eval { start_ls_files($self, $di) }; ERR($self, $@) if $@; @@ -409,17 +504,21 @@ sub di_url ($$) { sub resolve_patch ($$) { my ($self, $want) = @_; - if (scalar(@{$self->{patches}}) > $self->{max_patch}) { + if (scalar(@{$self->{patches}}) > $MAX_PATCH) { die "Aborting, too many steps to $self->{oid_want}"; } # see if we can find the blob in an existing git repo: my $cur_want = $want->{oid_b}; + if ($self->{seen_oid}->{$cur_want}++) { + die "Loop detected solving $cur_want\n"; + } if (my $existing = solve_existing($self, $want)) { + my ($found_git, undef, $type, undef) = @$existing; dbg($self, "found $cur_want in " . - join("\n", $existing->[0]->pub_urls)); + join("\n", $found_git->pub_urls($self->{psgi_env}))); - if ($cur_want eq $self->{oid_want}) { # all done! + if ($cur_want eq $self->{oid_want} || $type ne 'blob') { eval { delete($self->{user_cb})->($existing) }; die "E: $@" if $@; return; @@ -430,15 +529,16 @@ sub resolve_patch ($$) { # 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; + my $diffs = find_extract_diffs($self, $ibx, $want) or next; - unshift @{$self->{patches}}, $di; - dbg($self, "found $cur_want in ".di_url($self, $di)); + unshift @{$self->{patches}}, @$diffs; + dbg($self, "found $cur_want in ". + join("\n\t", map { di_url($self, $_) } @$diffs)); # 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 = $di->{oid_a}; unless ($src =~ /\A0+\z/) { @@ -448,6 +548,14 @@ sub resolve_patch ($$) { } return next_step($self); # onto the next todo item } + if (length($cur_want) > $OID_MIN) { + chop($cur_want); + dbg($self, "retrying $want->{oid_b} as $cur_want"); + $want->{oid_b} = $cur_want; + push @{$self->{todo}}, $want; + return next_step($self); # retry with shorter abbrev + } + dbg($self, "could not find $cur_want"); eval { delete($self->{user_cb})->(undef) }; # not found! :< die "E: $@" if $@; @@ -461,7 +569,6 @@ sub new { bless { gits => $ibx->{-repo_objs}, user_cb => $user_cb, - max_patch => 100, # TODO: config option for searching related inboxes inboxes => [ $ibx ], @@ -469,6 +576,7 @@ sub new { } # recreate $oid_want using $hints +# hints keys: path_a, path_b, oid_a # Calls {user_cb} with: [ ::Git object, oid_full, type, size, di (diff_info) ] # with found object, or undef if nothing was found # Calls {user_cb} with a string error on fatal errors @@ -481,11 +589,13 @@ sub solve ($$$$$) { $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.tmp-XXXXXXXX', TMPDIR => 1); + $self->{tmp} = File::Temp->newdir("solver.$oid_want-XXXXXXXX", TMPDIR => 1); dbg($self, "solving $oid_want ..."); my $step_cb = step_cb($self);