2 package IkiWiki::Plugin::git;
8 use open qw{:utf8 :std};
10 my $sha1_pattern = qr/[0-9a-fA-F]{40}/; # pattern to validate Git sha1sums
11 my $dummy_commit_msg = 'dummy commit'; # message to skip in recent changes
15 hook(type => "checkconfig", id => "git", call => \&checkconfig);
16 hook(type => "getsetup", id => "git", call => \&getsetup);
17 hook(type => "genwrapper", id => "git", call => \&genwrapper);
18 hook(type => "rcs", id => "rcs_update", call => \&rcs_update);
19 hook(type => "rcs", id => "rcs_prepedit", call => \&rcs_prepedit);
20 hook(type => "rcs", id => "rcs_commit", call => \&rcs_commit);
21 hook(type => "rcs", id => "rcs_commit_staged", call => \&rcs_commit_staged);
22 hook(type => "rcs", id => "rcs_add", call => \&rcs_add);
23 hook(type => "rcs", id => "rcs_remove", call => \&rcs_remove);
24 hook(type => "rcs", id => "rcs_rename", call => \&rcs_rename);
25 hook(type => "rcs", id => "rcs_recentchanges", call => \&rcs_recentchanges);
26 hook(type => "rcs", id => "rcs_diff", call => \&rcs_diff);
27 hook(type => "rcs", id => "rcs_getctime", call => \&rcs_getctime);
28 hook(type => "rcs", id => "rcs_getmtime", call => \&rcs_getmtime);
29 hook(type => "rcs", id => "rcs_receive", call => \&rcs_receive);
30 hook(type => "rcs", id => "rcs_preprevert", call => \&rcs_preprevert);
31 hook(type => "rcs", id => "rcs_revert", call => \&rcs_revert);
35 if (! defined $config{gitorigin_branch}) {
36 $config{gitorigin_branch}="origin";
38 if (! defined $config{gitmaster_branch}) {
39 $config{gitmaster_branch}="master";
41 if (defined $config{git_wrapper} &&
42 length $config{git_wrapper}) {
43 push @{$config{wrappers}}, {
44 wrapper => $config{git_wrapper},
45 wrappermode => (defined $config{git_wrappermode} ? $config{git_wrappermode} : "06755"),
46 wrapper_background_command => $config{git_wrapper_background_command},
50 if (defined $config{git_test_receive_wrapper} &&
51 length $config{git_test_receive_wrapper} &&
52 defined $config{untrusted_committers} &&
53 @{$config{untrusted_committers}}) {
54 push @{$config{wrappers}}, {
56 wrapper => $config{git_test_receive_wrapper},
57 wrappermode => (defined $config{git_wrappermode} ? $config{git_wrappermode} : "06755"),
61 # Avoid notes, parser does not handle and they only slow things down.
62 $ENV{GIT_NOTES_REF}="";
64 # Run receive test only if being called by the wrapper, and not
65 # when generating same.
66 if ($config{test_receive} && ! exists $config{wrapper}) {
67 require IkiWiki::Receive;
68 IkiWiki::Receive::test();
75 safe => 0, # rcs plugin
81 example => "/git/wiki.git/hooks/post-update",
82 description => "git hook to generate",
86 git_wrapper_background_command => {
88 example => "git push github",
89 description => "shell command for git_wrapper to run, in the background",
96 description => "mode for git_wrapper (can safely be made suid)",
100 git_test_receive_wrapper => {
102 example => "/git/wiki.git/hooks/pre-receive",
103 description => "git pre-receive hook to generate",
107 untrusted_committers => {
110 description => "unix users whose commits should be checked by the pre-receive hook",
116 example => "http://git.example.com/gitweb.cgi?p=wiki.git;a=history;f=[[file]];hb=HEAD",
117 description => "gitweb url to show file history ([[file]] substituted)",
123 example => "http://git.example.com/gitweb.cgi?p=wiki.git;a=blobdiff;f=[[file]];h=[[sha1_to]];hp=[[sha1_from]];hb=[[sha1_commit]];hpb=[[sha1_parent]]",
124 description => "gitweb url to show a diff ([[file]], [[sha1_to]], [[sha1_from]], [[sha1_commit]], and [[sha1_parent]] substituted)",
128 gitorigin_branch => {
131 description => "where to pull and push changes (set to empty string to disable)",
132 safe => 0, # paranoia
135 gitmaster_branch => {
138 description => "branch that the wiki is stored in",
139 safe => 0, # paranoia
145 if ($config{test_receive}) {
146 require IkiWiki::Receive;
147 return IkiWiki::Receive::genwrapper();
155 # Start a child process safely without resorting /bin/sh.
156 # Return command output or success state (in scalar context).
158 my ($error_handler, @cmdline) = @_;
160 my $pid = open my $OUT, "-|";
162 error("Cannot fork: $!") if !defined $pid;
166 # Git commands want to be in wc.
167 if (! defined $git_dir) {
168 chdir $config{srcdir}
169 or error("cannot chdir to $config{srcdir}: $!");
173 or error("cannot chdir to $git_dir: $!");
175 exec @cmdline or error("Cannot exec '@cmdline': $!");
179 # git output is probably utf-8 encoded, but may contain
180 # other encodings or invalidly encoded stuff. So do not rely
181 # on the normal utf-8 IO layer, decode it by hand.
186 $_=decode_utf8($_, 0);
195 $error_handler->("'@cmdline' failed: $!") if $? && $error_handler;
197 return wantarray ? @lines : ($? == 0);
199 # Convenient wrappers.
200 sub run_or_die ($@) { safe_git(\&error, @_) }
201 sub run_or_cry ($@) { safe_git(sub { warn @_ }, @_) }
202 sub run_or_non ($@) { safe_git(undef, @_) }
205 sub merge_past ($$$) {
206 # Unlike with Subversion, Git cannot make a 'svn merge -rN:M file'.
207 # Git merge commands work with the committed changes, except in the
208 # implicit case of '-m' of git checkout(1). So we should invent a
209 # kludge here. In principle, we need to create a throw-away branch
210 # in preparing for the merge itself. Since branches are cheap (and
211 # branching is fast), this shouldn't cost high.
213 # The main problem is the presence of _uncommitted_ local changes. One
214 # possible approach to get rid of this situation could be that we first
215 # make a temporary commit in the master branch and later restore the
216 # initial state (this is possible since Git has the ability to undo a
217 # commit, i.e. 'git reset --soft HEAD^'). The method can be summarized
220 # - create a diff of HEAD:current-sha1
222 # - create a dummy branch and switch to it
223 # - rewind to past (reset --hard to the current-sha1)
224 # - apply the diff and commit
225 # - switch to master and do the merge with the dummy branch
226 # - make a soft reset (undo the last commit of master)
228 # The above method has some drawbacks: (1) it needs a redundant commit
229 # just to get rid of local changes, (2) somewhat slow because of the
230 # required system forks. Until someone points a more straight method
231 # (which I would be grateful) I have implemented an alternative method.
232 # In this approach, we hide all the modified files from Git by renaming
233 # them (using the 'rename' builtin) and later restore those files in
234 # the throw-away branch (that is, we put the files themselves instead
235 # of applying a patch).
237 my ($sha1, $file, $message) = @_;
239 my @undo; # undo stack for cleanup in case of an error
240 my $conflict; # file content with conflict markers
243 # Hide local changes from Git by renaming the modified file.
244 # Relative paths must be converted to absolute for renaming.
245 my ($target, $hidden) = (
246 "$config{srcdir}/${file}", "$config{srcdir}/${file}.${sha1}"
248 rename($target, $hidden)
249 or error("rename '$target' to '$hidden' failed: $!");
250 # Ensure to restore the renamed file on error.
252 return if ! -e "$hidden"; # already renamed
253 rename($hidden, $target)
254 or warn "rename '$hidden' to '$target' failed: $!";
257 my $branch = "throw_away_${sha1}"; # supposed to be unique
259 # Create a throw-away branch and rewind backward.
260 push @undo, sub { run_or_cry('git', 'branch', '-D', $branch) };
261 run_or_die('git', 'branch', $branch, $sha1);
263 # Switch to throw-away branch for the merge operation.
265 if (!run_or_cry('git', 'checkout', $config{gitmaster_branch})) {
266 run_or_cry('git', 'checkout','-f',$config{gitmaster_branch});
269 run_or_die('git', 'checkout', $branch);
271 # Put the modified file in _this_ branch.
272 rename($hidden, $target)
273 or error("rename '$hidden' to '$target' failed: $!");
275 # _Silently_ commit all modifications in the current branch.
276 run_or_non('git', 'commit', '-m', $message, '-a');
277 # ... and re-switch to master.
278 run_or_die('git', 'checkout', $config{gitmaster_branch});
280 # Attempt to merge without complaining.
281 if (!run_or_non('git', 'pull', '--no-commit', '.', $branch)) {
282 $conflict = readfile($target);
283 run_or_die('git', 'reset', '--hard');
288 # Process undo stack (in reverse order). By policy cleanup
289 # actions should normally print a warning on failure.
290 while (my $handle = pop @undo) {
294 error("Git merge failed!\n$failure\n") if $failure;
301 sub decode_git_file ($) {
304 # git does not output utf-8 filenames, but instead
305 # double-quotes them with the utf-8 characters
306 # escaped as \nnn\nnn.
307 if ($file =~ m/^"(.*)"$/) {
308 ($file=$1) =~ s/\\([0-7]{1,3})/chr(oct($1))/eg;
311 # strip prefix if in a subdir
312 if (! defined $prefix) {
313 ($prefix) = run_or_die('git', 'rev-parse', '--show-prefix');
314 if (! defined $prefix) {
318 $file =~ s/^\Q$prefix\E//;
320 return decode("utf8", $file);
324 sub parse_diff_tree ($) {
325 # Parse the raw diff tree chunk and return the info hash.
326 # See git-diff-tree(1) for the syntax.
330 return if !defined @{ $dt_ref } ||
331 !defined @{ $dt_ref }[0] || !length @{ $dt_ref }[0];
335 while (my $line = shift @{ $dt_ref }) {
336 return if $line !~ m/^(.+) ($sha1_pattern)/;
343 # Identification lines for the commit.
344 while (my $line = shift @{ $dt_ref }) {
345 # Regexps are semi-stolen from gitweb.cgi.
346 if ($line =~ m/^tree ([0-9a-fA-F]{40})$/) {
349 elsif ($line =~ m/^parent ([0-9a-fA-F]{40})$/) {
350 # XXX: collecting in reverse order
351 push @{ $ci{'parents'} }, $1;
353 elsif ($line =~ m/^(author|committer) (.*) ([0-9]+) (.*)$/) {
354 my ($who, $name, $epoch, $tz) =
358 $ci{ "${who}_epoch" } = $epoch;
359 $ci{ "${who}_tz" } = $tz;
361 if ($name =~ m/^([^<]+)\s+<([^@>]+)/) {
362 $ci{"${who}_name"} = $1;
363 $ci{"${who}_username"} = $2;
365 elsif ($name =~ m/^([^<]+)\s+<>$/) {
366 $ci{"${who}_username"} = $1;
369 $ci{"${who}_username"} = $name;
372 elsif ($line =~ m/^$/) {
373 # Trailing empty line signals next section.
378 debug("No 'tree' seen in diff-tree output") if !defined $ci{'tree'};
380 if (defined $ci{'parents'}) {
381 $ci{'parent'} = @{ $ci{'parents'} }[0];
384 $ci{'parent'} = 0 x 40;
387 # Commit message (optional).
388 while ($dt_ref->[0] =~ /^ /) {
389 my $line = shift @{ $dt_ref };
391 push @{ $ci{'comment'} }, $line;
393 shift @{ $dt_ref } if $dt_ref->[0] =~ /^$/;
396 while (my $line = shift @{ $dt_ref }) {
398 (:+) # number of parents
399 ([^\t]+)\t # modes, sha1, status
402 my $num_parents = length $1;
403 my @tmp = split(" ", $2);
404 my ($file, $file_to) = split("\t", $3);
405 my @mode_from = splice(@tmp, 0, $num_parents);
406 my $mode_to = shift(@tmp);
407 my @sha1_from = splice(@tmp, 0, $num_parents);
408 my $sha1_to = shift(@tmp);
409 my $status = shift(@tmp);
412 push @{ $ci{'details'} }, {
413 'file' => decode_git_file($file),
414 'sha1_from' => $sha1_from[0],
415 'sha1_to' => $sha1_to,
416 'mode_from' => $mode_from[0],
417 'mode_to' => $mode_to,
429 sub git_commit_info ($;$) {
430 # Return an array of commit info hashes of num commits
431 # starting from the given sha1sum.
432 my ($sha1, $num) = @_;
435 push @opts, "--max-count=$num" if defined $num;
437 my @raw_lines = run_or_die('git', 'log', @opts,
438 '--pretty=raw', '--raw', '--abbrev=40', '--always', '-c',
439 '-r', $sha1, '--', '.');
442 while (my $parsed = parse_diff_tree(\@raw_lines)) {
446 warn "Cannot parse commit info for '$sha1' commit" if !@ci;
448 return wantarray ? @ci : $ci[0];
452 # Return head sha1sum (of given file).
453 my $file = shift || q{--};
455 # Ignore error since a non-existing file might be given.
456 my ($sha1) = run_or_non('git', 'rev-list', '--max-count=1', 'HEAD',
459 ($sha1) = $sha1 =~ m/($sha1_pattern)/; # sha1 is untainted now
462 debug("Empty sha1sum for '$file'.");
464 return defined $sha1 ? $sha1 : q{};
468 # Update working directory.
470 if (length $config{gitorigin_branch}) {
471 run_or_cry('git', 'pull', '--prune', $config{gitorigin_branch});
475 sub rcs_prepedit ($) {
476 # Return the commit sha1sum of the file when editing begins.
477 # This will be later used in rcs_commit if a merge is required.
480 return git_sha1($file);
484 # Try to commit the page; returns undef on _success_ and
485 # a version of the page with the rcs's conflict markers on
489 # Check to see if the page has been changed by someone else since
490 # rcs_prepedit was called.
491 my $cur = git_sha1($params{file});
492 my ($prev) = $params{token} =~ /^($sha1_pattern)$/; # untaint
494 if (defined $cur && defined $prev && $cur ne $prev) {
495 my $conflict = merge_past($prev, $params{file}, $dummy_commit_msg);
496 return $conflict if defined $conflict;
499 rcs_add($params{file});
500 return rcs_commit_staged(
501 message => $params{message},
502 session => $params{session},
506 sub rcs_commit_staged (@) {
507 # Commits all staged changes. Changes can be staged using rcs_add,
508 # rcs_remove, and rcs_rename.
513 if (defined $params{session}) {
514 # Set the commit author and email based on web session info.
516 if (defined $params{session}->param("name")) {
517 $u=$params{session}->param("name");
519 elsif (defined $params{session}->remote_addr()) {
520 $u=$params{session}->remote_addr();
524 $ENV{GIT_AUTHOR_NAME}=$u;
526 if (defined $params{session}->param("nickname")) {
527 $u=encode_utf8($params{session}->param("nickname"));
529 $u=~s/[^-_0-9[:alnum:]]+//g;
532 $ENV{GIT_AUTHOR_EMAIL}="$u\@web";
536 $params{message} = IkiWiki::possibly_foolish_untaint($params{message});
538 if ($params{message} !~ /\S/) {
539 # Force git to allow empty commit messages.
540 # (If this version of git supports it.)
541 my ($version)=`git --version` =~ /git version (.*)/;
542 if ($version ge "1.5.4") {
543 push @opts, '--cleanup=verbatim';
546 $params{message}.=".";
550 # git commit returns non-zero if file has not been really changed.
551 # so we should ignore its exit status (hence run_or_non).
552 if (run_or_non('git', 'commit', @opts, '-m', $params{message})) {
553 if (length $config{gitorigin_branch}) {
554 run_or_cry('git', 'push', $config{gitorigin_branch});
559 return undef; # success
563 # Add file to archive.
567 run_or_cry('git', 'add', $file);
571 # Remove file from archive.
575 run_or_cry('git', 'rm', '-f', $file);
578 sub rcs_rename ($$) {
579 my ($src, $dest) = @_;
581 run_or_cry('git', 'mv', '-f', $src, $dest);
584 sub rcs_recentchanges ($) {
585 # List of recent changes.
589 eval q{use Date::Parse};
593 foreach my $ci (git_commit_info('HEAD', $num || 1)) {
594 # Skip redundant commits.
595 next if ($ci->{'comment'} && @{$ci->{'comment'}}[0] eq $dummy_commit_msg);
597 my ($sha1, $when) = (
599 $ci->{'author_epoch'}
603 foreach my $detail (@{ $ci->{'details'} }) {
604 my $file = $detail->{'file'};
606 my $diffurl = defined $config{'diffurl'} ? $config{'diffurl'} : "";
607 $diffurl =~ s/\[\[file\]\]/$file/go;
608 $diffurl =~ s/\[\[sha1_parent\]\]/$ci->{'parent'}/go;
609 $diffurl =~ s/\[\[sha1_from\]\]/$detail->{'sha1_from'}/go;
610 $diffurl =~ s/\[\[sha1_to\]\]/$detail->{'sha1_to'}/go;
611 $diffurl =~ s/\[\[sha1_commit\]\]/$sha1/go;
614 page => pagename($file),
621 foreach my $line (@{$ci->{'comment'}}) {
622 $pastblank=1 if $line eq '';
623 next if $pastblank && $line=~m/^ *(signed[ \-]off[ \-]by[ :]|acked[ \-]by[ :]|cc[ :])/i;
624 push @messages, { line => $line };
627 my $user=$ci->{'author_username'};
628 my $web_commit = ($ci->{'author'} =~ /\@web>/);
631 # Set nickname only if a non-url author_username is available,
632 # and author_name is an url.
633 if ($user !~ /:\/\// && defined $ci->{'author_name'} &&
634 $ci->{'author_name'} =~ /:\/\//) {
636 $user=$ci->{'author_name'};
639 # compatability code for old web commit messages
641 defined $messages[0] &&
642 $messages[0]->{line} =~ m/$config{web_commit_regexp}/) {
643 $user = defined $2 ? "$2" : "$3";
644 $messages[0]->{line} = $4;
651 nickname => $nickname,
652 committype => $web_commit ? "web" : "git",
654 message => [@messages],
658 last if @rets >= $num;
666 my ($sha1) = $rev =~ /^($sha1_pattern)$/; # untaint
668 foreach my $line (run_or_non("git", "show", $sha1)) {
669 if (@lines || $line=~/^diff --git/) {
670 push @lines, $line."\n";
677 return join("", @lines);
686 my $id=shift; # 0 = mtime ; 1 = ctime
688 if (! keys %time_cache) {
690 foreach my $line (run_or_die('git', 'log',
691 '--pretty=format:%ct',
692 '--name-only', '--relative')) {
693 if (! defined $date && $line =~ /^(\d+)$/) {
696 elsif (! length $line) {
700 my $f=decode_git_file($line);
702 if (! $time_cache{$f}) {
703 $time_cache{$f}[0]=$date; # mtime
705 $time_cache{$f}[1]=$date; # ctime
710 return exists $time_cache{$file} ? $time_cache{$file}[$id] : 0;
715 sub rcs_getctime ($) {
718 return findtimes($file, 1);
721 sub rcs_getmtime ($) {
724 return findtimes($file, 0);
730 # The wiki may not be the only thing in the git repo.
731 # Determine if it is in a subdirectory by examining the srcdir,
732 # and its parents, looking for the .git directory.
734 return @$ret if defined $ret;
737 my $dir=$config{srcdir};
738 while (! -d "$dir/.git") {
739 $subdir=IkiWiki::basename($dir)."/".$subdir;
740 $dir=IkiWiki::dirname($dir);
742 error("cannot determine root of git repo");
746 $ret=[$subdir, $dir];
752 sub git_parse_changes {
755 my ($subdir, $rootdir) = git_find_root();
757 foreach my $ci (@changes) {
758 foreach my $detail (@{ $ci->{'details'} }) {
759 my $file = $detail->{'file'};
761 # check that all changed files are in the subdir
762 if (length $subdir &&
763 ! ($file =~ s/^\Q$subdir\E//)) {
764 error sprintf(gettext("you are not allowed to change %s"), $file);
767 my ($action, $mode, $path);
768 if ($detail->{'status'} =~ /^[M]+\d*$/) {
770 $mode=$detail->{'mode_to'};
772 elsif ($detail->{'status'} =~ /^[AM]+\d*$/) {
774 $mode=$detail->{'mode_to'};
776 elsif ($detail->{'status'} =~ /^[DAM]+\d*/) {
778 $mode=$detail->{'mode_from'};
781 error "unknown status ".$detail->{'status'};
784 # test that the file mode is ok
785 if ($mode !~ /^100[64][64][64]$/) {
786 error sprintf(gettext("you cannot act on a file with mode %s"), $mode);
788 if ($action eq "change") {
789 if ($detail->{'mode_from'} ne $detail->{'mode_to'}) {
790 error gettext("you are not allowed to change file modes");
794 # extract attachment to temp file
795 if (($action eq 'add' || $action eq 'change') &&
797 eval q{use File::Temp};
800 ($fh, $path)=File::Temp::tempfile("XXXXXXXXXX", UNLINK => 1);
801 my $cmd = "cd $git_dir && ".
802 "git show $detail->{sha1_to} > '$path'";
803 if (system($cmd) != 0) {
804 error("failed writing temp file '$path'.");
823 my ($oldrev, $newrev, $refname) = split(' ', $_, 3);
825 # only allow changes to gitmaster_branch
826 if ($refname !~ /^refs\/heads\/\Q$config{gitmaster_branch}\E$/) {
827 error sprintf(gettext("you are not allowed to change %s"), $refname);
830 # Avoid chdir when running git here, because the changes
831 # are in the master git repo, not the srcdir repo.
832 # (Also, if a subdir is involved, we don't want to chdir to
833 # it and only see changes in it.)
834 # The pre-receive hook already puts us in the right place.
836 push @rets, git_parse_changes(git_commit_info($oldrev."..".$newrev));
840 return reverse @rets;
843 sub rcs_preprevert ($) {
845 my ($sha1) = $rev =~ /^($sha1_pattern)$/; # untaint
847 # Examine changes from root of git repo, not from any subdir,
848 # in order to see all changes.
849 my ($subdir, $rootdir) = git_find_root();
851 my @commits=git_commit_info($sha1, 1);
855 error "unknown commit"; # just in case
858 # git revert will fail on merge commits. Add a nice message.
859 if (exists $commits[0]->{parents} &&
860 @{$commits[0]->{parents}} > 1) {
861 error gettext("you are not allowed to revert a merge");
864 return git_parse_changes(@commits);
868 # Try to revert the given rev; returns undef on _success_.
870 my ($sha1) = $rev =~ /^($sha1_pattern)$/; # untaint
872 if (run_or_non('git', 'revert', '--no-commit', $sha1)) {
876 run_or_die('git', 'reset', '--hard');
877 return sprintf(gettext("Failed to revert commit %s"), $sha1);