contrib/git-svn: stabilize memory usage for big fetches
[git.git] / contrib / git-svn / git-svn.perl
index a32ce15..3e5733e 100755 (executable)
@@ -4,19 +4,12 @@
 use warnings;
 use strict;
 use vars qw/   $AUTHOR $VERSION
-               $SVN_URL $SVN_INFO $SVN_WC
+               $SVN_URL $SVN_INFO $SVN_WC $SVN_UUID
                $GIT_SVN_INDEX $GIT_SVN
                $GIT_DIR $REV_DIR/;
 $AUTHOR = 'Eric Wong <normalperson@yhbt.net>';
-$VERSION = '0.9.1';
+$VERSION = '0.11.0';
 $GIT_DIR = $ENV{GIT_DIR} || "$ENV{PWD}/.git";
-$GIT_SVN = $ENV{GIT_SVN_ID} || 'git-svn';
-$GIT_SVN_INDEX = "$GIT_DIR/$GIT_SVN/index";
-$ENV{GIT_DIR} ||= $GIT_DIR;
-$SVN_URL = undef;
-$REV_DIR = "$GIT_DIR/$GIT_SVN/revs";
-$SVN_WC = "$GIT_DIR/$GIT_SVN/tree";
-
 # make sure the svn binary gives consistent output between locales and TZs:
 $ENV{TZ} = 'UTC';
 $ENV{LC_ALL} = 'C';
@@ -24,33 +17,40 @@ $ENV{LC_ALL} = 'C';
 # If SVN:: library support is added, please make the dependencies
 # optional and preserve the capability to use the command-line client.
 # use eval { require SVN::... } to make it lazy load
+# We don't use any modules not in the standard Perl distribution:
 use Carp qw/croak/;
 use IO::File qw//;
 use File::Basename qw/dirname basename/;
 use File::Path qw/mkpath/;
 use Getopt::Long qw/:config gnu_getopt no_ignore_case auto_abbrev/;
 use File::Spec qw//;
+use POSIX qw/strftime/;
 my $sha1 = qr/[a-f\d]{40}/;
-my $sha1_short = qr/[a-f\d]{6,40}/;
+my $sha1_short = qr/[a-f\d]{4,40}/;
 my ($_revision,$_stdin,$_no_ignore_ext,$_no_stop_copy,$_help,$_rmdir,$_edit,
-       $_find_copies_harder, $_l, $_version);
-
-GetOptions(    'revision|r=s' => \$_revision,
-               'no-ignore-externals' => \$_no_ignore_ext,
-               'stdin|' => \$_stdin,
-               'edit|e' => \$_edit,
-               'rmdir' => \$_rmdir,
-               'help|H|h' => \$_help,
-               'find-copies-harder' => \$_find_copies_harder,
-               'l=i' => \$_l,
-               'version|V' => \$_version,
-               'no-stop-on-copy' => \$_no_stop_copy );
+       $_find_copies_harder, $_l, $_version, $_upgrade, $_authors);
+my (@_branch_from, %tree_map, %users);
+my $_svn_co_url_revs;
+
+my %fc_opts = ( 'no-ignore-externals' => \$_no_ignore_ext,
+               'branch|b=s' => \@_branch_from,
+               'authors-file|A=s' => \$_authors );
 my %cmd = (
-       fetch => [ \&fetch, "Download new revisions from SVN" ],
-       init => [ \&init, "Initialize and fetch (import)"],
-       commit => [ \&commit, "Commit git revisions to SVN" ],
-       rebuild => [ \&rebuild, "Rebuild git-svn metadata (after git clone)" ],
-       help => [ \&usage, "Show help" ],
+       fetch => [ \&fetch, "Download new revisions from SVN",
+                       { 'revision|r=s' => \$_revision, %fc_opts } ],
+       init => [ \&init, "Initialize and fetch (import)", { } ],
+       commit => [ \&commit, "Commit git revisions to SVN",
+                       {       'stdin|' => \$_stdin,
+                               'edit|e' => \$_edit,
+                               'rmdir' => \$_rmdir,
+                               'find-copies-harder' => \$_find_copies_harder,
+                               'l=i' => \$_l,
+                               %fc_opts,
+                       } ],
+       'show-ignore' => [ \&show_ignore, "Show svn:ignore listings", { } ],
+       rebuild => [ \&rebuild, "Rebuild git-svn metadata (after git clone)",
+                       { 'no-ignore-externals' => \$_no_ignore_ext,
+                         'upgrade' => \$_upgrade } ],
 );
 my $cmd;
 for (my $i = 0; $i < @ARGV; $i++) {
@@ -61,17 +61,24 @@ for (my $i = 0; $i < @ARGV; $i++) {
        }
 };
 
-# we may be called as git-svn-(command), or git-svn(command).
-foreach (keys %cmd) {
-       if (/git\-svn\-?($_)(?:\.\w+)?$/) {
-               $cmd = $1;
-               last;
-       }
-}
+my %opts = %{$cmd{$cmd}->[2]} if (defined $cmd);
+
+GetOptions(%opts, 'help|H|h' => \$_help,
+               'version|V' => \$_version,
+               'id|i=s' => \$GIT_SVN) or exit 1;
+
+$GIT_SVN ||= $ENV{GIT_SVN_ID} || 'git-svn';
+$GIT_SVN_INDEX = "$GIT_DIR/$GIT_SVN/index";
+$ENV{GIT_DIR} ||= $GIT_DIR;
+$SVN_URL = undef;
+$REV_DIR = "$GIT_DIR/$GIT_SVN/revs";
+$SVN_WC = "$GIT_DIR/$GIT_SVN/tree";
+
 usage(0) if $_help;
 version() if $_version;
-usage(1) unless (defined $cmd);
-svn_check_ignore_externals();
+usage(1) unless defined $cmd;
+load_authors() if $_authors;
+svn_compat_check();
 $cmd{$cmd}->[0]->(@ARGV);
 exit 0;
 
@@ -82,15 +89,25 @@ sub usage {
        print $fd <<"";
 git-svn - bidirectional operations between a single Subversion tree and git
 Usage: $0 <command> [options] [arguments]\n
-Available commands:
+
+       print $fd "Available commands:\n" unless $cmd;
 
        foreach (sort keys %cmd) {
-               print $fd '  ',pack('A10',$_),$cmd{$_}->[1],"\n";
+               next if $cmd && $cmd ne $_;
+               print $fd '  ',pack('A13',$_),$cmd{$_}->[1],"\n";
+               foreach (keys %{$cmd{$_}->[2]}) {
+                       # prints out arguments as they should be passed:
+                       my $x = s#=s$## ? '<arg>' : s#=i$## ? '<num>' : '';
+                       print $fd ' ' x 17, join(', ', map { length $_ > 1 ?
+                                                       "--$_" : "-$_" }
+                                               split /\|/,$_)," $x\n";
+               }
        }
        print $fd <<"";
-\nGIT_SVN_ID may be set in the environment to an arbitrary identifier if
-you're tracking multiple SVN branches/repositories in one git repository
-and want to keep them separate.
+\nGIT_SVN_ID may be set in the environment or via the --id/-i switch to an
+arbitrary identifier if you're tracking multiple SVN branches/repositories in
+one git repository and want to keep them separate.  See git-svn(1) for more
+information.
 
        exit $exit;
 }
@@ -102,15 +119,19 @@ sub version {
 
 sub rebuild {
        $SVN_URL = shift or undef;
-       my $repo_uuid;
        my $newest_rev = 0;
+       if ($_upgrade) {
+               sys('git-update-ref',"refs/remotes/$GIT_SVN","$GIT_SVN-HEAD");
+       } else {
+               check_upgrade_needed();
+       }
 
        my $pid = open(my $rev_list,'-|');
        defined $pid or croak $!;
        if ($pid == 0) {
-               exec("git-rev-list","$GIT_SVN-HEAD") or croak $!;
+               exec("git-rev-list","refs/remotes/$GIT_SVN") or croak $!;
        }
-       my $first;
+       my $latest;
        while (<$rev_list>) {
                chomp;
                my $c = $_;
@@ -130,18 +151,21 @@ sub rebuild {
                                        "$c, $id\n";
                        }
                }
+
+               # if we merged or otherwise started elsewhere, this is
+               # how we break out of it
+               next if (defined $SVN_UUID && ($uuid ne $SVN_UUID));
+               next if (defined $SVN_URL && defined $url && ($url ne $SVN_URL));
+
                print "r$rev = $c\n";
-               unless (defined $first) {
+               unless (defined $latest) {
                        if (!$SVN_URL && !$url) {
                                croak "SVN repository location required: $url\n";
                        }
                        $SVN_URL ||= $url;
-                       $repo_uuid = setup_git_svn();
-                       $first = $rev;
-               }
-               if ($uuid ne $repo_uuid) {
-                       croak "Repository UUIDs do not match!\ngot: $uuid\n",
-                                               "expected: $repo_uuid\n";
+                       $SVN_UUID ||= $uuid;
+                       setup_git_svn();
+                       $latest = $rev;
                }
                assert_revision_eq_or_unknown($rev, $c);
                sys('git-update-ref',"$GIT_SVN/revs/$rev",$c);
@@ -149,9 +173,7 @@ sub rebuild {
        }
        close $rev_list or croak $?;
        if (!chdir $SVN_WC) {
-               my @svn_co = ('svn','co',"-r$first");
-               push @svn_co, '--ignore-externals' unless $_no_ignore_ext;
-               sys(@svn_co, $SVN_URL, $SVN_WC);
+               svn_cmd_checkout($SVN_URL, $latest, $SVN_WC);
                chdir $SVN_WC or croak $!;
        }
 
@@ -166,6 +188,13 @@ sub rebuild {
                exec('git-write-tree');
        }
        waitpid $pid, 0;
+
+       if ($_upgrade) {
+               print STDERR <<"";
+Keeping deprecated refs/head/$GIT_SVN-HEAD for now.  Please remove it
+when you have upgraded your tools and habits to use refs/remotes/$GIT_SVN
+
+       }
 }
 
 sub init {
@@ -178,6 +207,7 @@ sub init {
 
 sub fetch {
        my (@parents) = @_;
+       check_upgrade_needed();
        $SVN_URL ||= file_to_s("$GIT_DIR/$GIT_SVN/info/url");
        my @log_args = -d $SVN_WC ? ($SVN_WC) : ($SVN_URL);
        unless ($_revision) {
@@ -187,44 +217,48 @@ sub fetch {
        push @log_args, '--stop-on-copy' unless $_no_stop_copy;
 
        my $svn_log = svn_log_raw(@log_args);
-       @$svn_log = sort { $a->{revision} <=> $b->{revision} } @$svn_log;
 
-       my $base = shift @$svn_log or croak "No base revision!\n";
+       my $base = next_log_entry($svn_log) or croak "No base revision!\n";
        my $last_commit = undef;
        unless (-d $SVN_WC) {
-               my @svn_co = ('svn','co',"-r$base->{revision}");
-               push @svn_co,'--ignore-externals' unless $_no_ignore_ext;
-               sys(@svn_co, $SVN_URL, $SVN_WC);
+               svn_cmd_checkout($SVN_URL,$base->{revision},$SVN_WC);
                chdir $SVN_WC or croak $!;
+               read_uuid();
                $last_commit = git_commit($base, @parents);
-               unless (-f "$GIT_DIR/refs/heads/master") {
-                       sys(qw(git-update-ref refs/heads/master),$last_commit);
-               }
                assert_svn_wc_clean($base->{revision}, $last_commit);
        } else {
                chdir $SVN_WC or croak $!;
+               read_uuid();
                $last_commit = file_to_s("$REV_DIR/$base->{revision}");
        }
        my @svn_up = qw(svn up);
        push @svn_up, '--ignore-externals' unless $_no_ignore_ext;
-       my $last_rev = $base->{revision};
-       foreach my $log_msg (@$svn_log) {
-               assert_svn_wc_clean($last_rev, $last_commit);
-               $last_rev = $log_msg->{revision};
-               sys(@svn_up,"-r$last_rev");
+       my $last = $base;
+       while (my $log_msg = next_log_entry($svn_log)) {
+               assert_svn_wc_clean($last->{revision}, $last_commit);
+               if ($last->{revision} >= $log_msg->{revision}) {
+                       croak "Out of order: last >= current: ",
+                               "$last->{revision} >= $log_msg->{revision}\n";
+               }
+               sys(@svn_up,"-r$log_msg->{revision}");
                $last_commit = git_commit($log_msg, $last_commit, @parents);
+               $last = $log_msg;
        }
-       assert_svn_wc_clean($last_rev, $last_commit);
-       return pop @$svn_log;
+       assert_svn_wc_clean($last->{revision}, $last_commit);
+       unless (-e "$GIT_DIR/refs/heads/master") {
+               sys(qw(git-update-ref refs/heads/master),$last_commit);
+       }
+       return $last;
 }
 
 sub commit {
        my (@commits) = @_;
+       check_upgrade_needed();
        if ($_stdin || !@commits) {
                print "Reading from stdin...\n";
                @commits = ();
                while (<STDIN>) {
-                       if (/\b([a-f\d]{6,40})\b/) {
+                       if (/\b($sha1_short)\b/o) {
                                unshift @commits, $1;
                        }
                }
@@ -244,9 +278,10 @@ sub commit {
 
        fetch();
        chdir $SVN_WC or croak $!;
-       my $svn_current_rev =  svn_info('.')->{'Last Changed Rev'};
+       my $info = svn_info('.');
+       read_uuid($info);
+       my $svn_current_rev =  $info->{'Last Changed Rev'};
        foreach my $c (@revs) {
-               print "Committing $c\n";
                my $mods = svn_checkout_tree($svn_current_rev, $c);
                if (scalar @$mods == 0) {
                        print "Skipping, no changes detected\n";
@@ -258,8 +293,40 @@ sub commit {
 
 }
 
+sub show_ignore {
+       require File::Find or die $!;
+       my $exclude_file = "$GIT_DIR/info/exclude";
+       open my $fh, '<', $exclude_file or croak $!;
+       chomp(my @excludes = (<$fh>));
+       close $fh or croak $!;
+
+       $SVN_URL ||= file_to_s("$GIT_DIR/$GIT_SVN/info/url");
+       chdir $SVN_WC or croak $!;
+       my %ign;
+       File::Find::find({wanted=>sub{if(lstat $_ && -d _ && -d "$_/.svn"){
+               s#^\./##;
+               @{$ign{$_}} = safe_qx(qw(svn propget svn:ignore),$_);
+               }}, no_chdir=>1},'.');
+
+       print "\n# /\n";
+       foreach (@{$ign{'.'}}) { print '/',$_ if /\S/ }
+       delete $ign{'.'};
+       foreach my $i (sort keys %ign) {
+               print "\n# ",$i,"\n";
+               foreach (@{$ign{$i}}) { print '/',$i,'/',$_ if /\S/ }
+       }
+}
+
 ########################### utility functions #########################
 
+sub read_uuid {
+       return if $SVN_UUID;
+       my $info = shift || svn_info('.');
+       $SVN_UUID = $info->{'Repository UUID'} or
+                                       croak "Repository UUID unreadable\n";
+       s_to_file($SVN_UUID,"$GIT_DIR/$GIT_SVN/info/uuid");
+}
+
 sub setup_git_svn {
        defined $SVN_URL or croak "SVN repository location required\n";
        unless (-d $GIT_DIR) {
@@ -269,24 +336,27 @@ sub setup_git_svn {
        mkpath(["$GIT_DIR/$GIT_SVN/info"]);
        mkpath([$REV_DIR]);
        s_to_file($SVN_URL,"$GIT_DIR/$GIT_SVN/info/url");
-       my $uuid = svn_info($SVN_URL)->{'Repository UUID'} or
-                                       croak "Repository UUID unreadable\n";
-       s_to_file($uuid,"$GIT_DIR/$GIT_SVN/info/uuid");
 
        open my $fd, '>>', "$GIT_DIR/$GIT_SVN/info/exclude" or croak $!;
        print $fd '.svn',"\n";
        close $fd or croak $!;
-       return $uuid;
 }
 
 sub assert_svn_wc_clean {
        my ($svn_rev, $treeish) = @_;
        croak "$svn_rev is not an integer!\n" unless ($svn_rev =~ /^\d+$/);
        croak "$treeish is not a sha1!\n" unless ($treeish =~ /^$sha1$/o);
-       my $svn_info = svn_info('.');
-       if ($svn_rev != $svn_info->{'Last Changed Rev'}) {
-               croak "Expected r$svn_rev, got r",
-                               $svn_info->{'Last Changed Rev'},"\n";
+       my $lcr = svn_info('.')->{'Last Changed Rev'};
+       if ($svn_rev != $lcr) {
+               print STDERR "Checking for copy-tree ... ";
+               # use
+               my @diff = grep(/^Index: /,(safe_qx(qw(svn diff),
+                                               "-r$lcr:$svn_rev")));
+               if (@diff) {
+                       croak "Nope!  Expected r$svn_rev, got r$lcr\n";
+               } else {
+                       print STDERR "OK!\n";
+               }
        }
        my @status = grep(!/^Performing status on external/,(`svn status`));
        @status = grep(!/^\s*$/,@status);
@@ -469,7 +539,7 @@ sub svn_checkout_tree {
        my ($svn_rev, $treeish) = @_;
        my $from = file_to_s("$REV_DIR/$svn_rev");
        assert_svn_wc_clean($svn_rev,$from);
-       print "diff-tree '$from' '$treeish'\n";
+       print "diff-tree $from $treeish\n";
        my $pid = open my $diff_fh, '-|';
        defined $pid or croak $!;
        if ($pid == 0) {
@@ -480,7 +550,7 @@ sub svn_checkout_tree {
        }
        my $mods = parse_diff_tree($diff_fh);
        unless (@$mods) {
-               # git can do empty commits, SVN doesn't allow it...
+               # git can do empty commits, but SVN doesn't allow it...
                return $mods;
        }
        my ($rm, $add) = precommit_check($mods);
@@ -566,7 +636,8 @@ sub handle_rmdir {
 sub svn_commit_tree {
        my ($svn_rev, $commit) = @_;
        my $commit_msg = "$GIT_DIR/$GIT_SVN/.svn-commit.tmp.$$";
-       open my $msg, '>', $commit_msg  or croak $!;
+       my %log_msg = ( msg => '' );
+       open my $msg, '>', $commit_msg or croak $!;
 
        chomp(my $type = `git-cat-file -t $commit`);
        if ($type eq 'commit') {
@@ -580,6 +651,9 @@ sub svn_commit_tree {
                while (<$msg_fh>) {
                        if (!$in_msg) {
                                $in_msg = 1 if (/^\s*$/);
+                       } elsif (/^git-svn-id: /) {
+                               # skip this, we regenerate the correct one
+                               # on re-fetch anyways
                        } else {
                                print $msg $_ or croak $!;
                        }
@@ -592,6 +666,15 @@ sub svn_commit_tree {
                my $editor = $ENV{VISUAL} || $ENV{EDITOR} || 'vi';
                system($editor, $commit_msg);
        }
+
+       # file_to_s removes all trailing newlines, so just use chomp() here:
+       open $msg, '<', $commit_msg or croak $!;
+       { local $/; chomp($log_msg{msg} = <$msg>); }
+       close $msg or croak $!;
+
+       my ($oneline) = ($log_msg{msg} =~ /([^\n\r]+)/);
+       print "Committing $commit: $oneline\n";
+
        my @ci_output = safe_qx(qw(svn commit -F),$commit_msg);
        my ($committed) = grep(/^Committed revision \d+\./,@ci_output);
        unlink $commit_msg;
@@ -600,56 +683,89 @@ sub svn_commit_tree {
                        join("\n",@ci_output),"\n";
        my ($rev_committed) = ($committed =~ /^Committed revision (\d+)\./);
 
-       # resync immediately
-       my @svn_up = (qw(svn up), "-r$svn_rev");
+       my @svn_up = qw(svn up);
        push @svn_up, '--ignore-externals' unless $_no_ignore_ext;
+       if ($rev_committed == ($svn_rev + 1)) {
+               push @svn_up, "-r$rev_committed";
+               sys(@svn_up);
+               my $info = svn_info('.');
+               my $date = $info->{'Last Changed Date'} or die "Missing date\n";
+               if ($info->{'Last Changed Rev'} != $rev_committed) {
+                       croak "$info->{'Last Changed Rev'} != $rev_committed\n"
+               }
+               my ($Y,$m,$d,$H,$M,$S,$tz) = ($date =~
+                                       /(\d{4})\-(\d\d)\-(\d\d)\s
+                                        (\d\d)\:(\d\d)\:(\d\d)\s([\-\+]\d+)/x)
+                                        or croak "Failed to parse date: $date\n";
+               $log_msg{date} = "$tz $Y-$m-$d $H:$M:$S";
+               $log_msg{author} = $info->{'Last Changed Author'};
+               $log_msg{revision} = $rev_committed;
+               $log_msg{msg} .= "\n";
+               my $parent = file_to_s("$REV_DIR/$svn_rev");
+               git_commit(\%log_msg, $parent, $commit);
+               return $rev_committed;
+       }
+       # resync immediately
+       push @svn_up, "-r$svn_rev";
        sys(@svn_up);
        return fetch("$rev_committed=$commit")->{revision};
 }
 
+# read the entire log into a temporary file (which is removed ASAP)
+# and store the file handle + parser state
 sub svn_log_raw {
        my (@log_args) = @_;
-       my $pid = open my $log_fh,'-|';
+       my $log_fh = IO::File->new_tmpfile or croak $!;
+       my $pid = fork;
        defined $pid or croak $!;
-
-       if ($pid == 0) {
+       if (!$pid) {
+               open STDOUT, '>&', $log_fh or croak $!;
                exec (qw(svn log), @log_args) or croak $!
        }
+       waitpid $pid, 0;
+       croak if $?;
+       seek $log_fh, 0, 0 or croak $!;
+       return { state => 'sep', fh => $log_fh };
+}
 
-       my @svn_log;
-       my $state = 'sep';
-       while (<$log_fh>) {
+sub next_log_entry {
+       my $log = shift; # retval of svn_log_raw()
+       my $ret = undef;
+       my $fh = $log->{fh};
+
+       while (<$fh>) {
                chomp;
                if (/^\-{72}$/) {
-                       if ($state eq 'msg') {
-                               if ($svn_log[$#svn_log]->{lines}) {
-                                       $svn_log[$#svn_log]->{msg} .= $_."\n";
-                                       unless(--$svn_log[$#svn_log]->{lines}) {
-                                               $state = 'sep';
+                       if ($log->{state} eq 'msg') {
+                               if ($ret->{lines}) {
+                                       $ret->{msg} .= $_."\n";
+                                       unless(--$ret->{lines}) {
+                                               $log->{state} = 'sep';
                                        }
                                } else {
                                        croak "Log parse error at: $_\n",
-                                               $svn_log[$#svn_log]->{revision},
+                                               $ret->{revision},
                                                "\n";
                                }
                                next;
                        }
-                       if ($state ne 'sep') {
+                       if ($log->{state} ne 'sep') {
                                croak "Log parse error at: $_\n",
-                                       "state: $state\n",
-                                       $svn_log[$#svn_log]->{revision},
+                                       "state: $log->{state}\n",
+                                       $ret->{revision},
                                        "\n";
                        }
-                       $state = 'rev';
+                       $log->{state} = 'rev';
 
                        # if we have an empty log message, put something there:
-                       if (@svn_log) {
-                               $svn_log[$#svn_log]->{msg} ||= "\n";
-                               delete $svn_log[$#svn_log]->{lines};
+                       if ($ret) {
+                               $ret->{msg} ||= "\n";
+                               delete $ret->{lines};
+                               return $ret;
                        }
                        next;
                }
-               if ($state eq 'rev' && s/^r(\d+)\s*\|\s*//) {
+               if ($log->{state} eq 'rev' && s/^r(\d+)\s*\|\s*//) {
                        my $rev = $1;
                        my ($author, $date, $lines) = split(/\s*\|\s*/, $_, 3);
                        ($lines) = ($lines =~ /(\d+)/);
@@ -657,32 +773,34 @@ sub svn_log_raw {
                                        /(\d{4})\-(\d\d)\-(\d\d)\s
                                         (\d\d)\:(\d\d)\:(\d\d)\s([\-\+]\d+)/x)
                                         or croak "Failed to parse date: $date\n";
-                       my %log_msg = ( revision => $rev,
+                       $ret = {        revision => $rev,
                                        date => "$tz $Y-$m-$d $H:$M:$S",
                                        author => $author,
                                        lines => $lines,
-                                       msg => '' );
-                       push @svn_log, \%log_msg;
-                       $state = 'msg_start';
+                                       msg => '' };
+                       if (defined $_authors && ! defined $users{$author}) {
+                               die "Author: $author not defined in ",
+                                               "$_authors file\n";
+                       }
+                       $log->{state} = 'msg_start';
                        next;
                }
                # skip the first blank line of the message:
-               if ($state eq 'msg_start' && /^$/) {
-                       $state = 'msg';
-               } elsif ($state eq 'msg') {
-                       if ($svn_log[$#svn_log]->{lines}) {
-                               $svn_log[$#svn_log]->{msg} .= $_."\n";
-                               unless (--$svn_log[$#svn_log]->{lines}) {
-                                       $state = 'sep';
+               if ($log->{state} eq 'msg_start' && /^$/) {
+                       $log->{state} = 'msg';
+               } elsif ($log->{state} eq 'msg') {
+                       if ($ret->{lines}) {
+                               $ret->{msg} .= $_."\n";
+                               unless (--$ret->{lines}) {
+                                       $log->{state} = 'sep';
                                }
                        } else {
                                croak "Log parse error at: $_\n",
-                                       $svn_log[$#svn_log]->{revision},"\n";
+                                       $ret->{revision},"\n";
                        }
                }
        }
-       close $log_fh or croak $?;
-       return \@svn_log;
+       return $ret;
 }
 
 sub svn_info {
@@ -699,7 +817,7 @@ sub svn_info {
        # only single-lines seem to exist in svn info output
        while (<$info_fh>) {
                chomp $_;
-               if (m#^([^:]+)\s*:\s*(\S*)$#) {
+               if (m#^([^:]+)\s*:\s*(\S.*)$#) {
                        $ret->{$1} = $2;
                        push @{$ret->{-order}}, $1;
                }
@@ -745,11 +863,23 @@ sub assert_revision_unknown {
        }
 }
 
+sub trees_eq {
+       my ($x, $y) = @_;
+       my @x = safe_qx('git-cat-file','commit',$x);
+       my @y = safe_qx('git-cat-file','commit',$y);
+       if (($y[0] ne $x[0]) || $x[0] !~ /^tree $sha1\n$/
+                               || $y[0] !~ /^tree $sha1\n$/) {
+               print STDERR "Trees not equal: $y[0] != $x[0]\n";
+               return 0
+       }
+       return 1;
+}
+
 sub assert_revision_eq_or_unknown {
        my ($revno, $commit) = @_;
        if (-f "$REV_DIR/$revno") {
                my $current = file_to_s("$REV_DIR/$revno");
-               if ($commit ne $current) {
+               if (($commit ne $current) && !trees_eq($commit, $current)) {
                        croak "$REV_DIR/$revno already exists!\n",
                                "current: $current\nexpected: $commit\n";
                }
@@ -761,9 +891,8 @@ sub git_commit {
        my ($log_msg, @parents) = @_;
        assert_revision_unknown($log_msg->{revision});
        my $out_fh = IO::File->new_tmpfile or croak $!;
-       my $info = svn_info('.');
-       my $uuid = $info->{'Repository UUID'};
-       defined $uuid or croak "Unable to get Repository UUID\n";
+
+       map_tree_joins() if (@_branch_from && !%tree_map);
 
        # commit parents can be conditionally bound to a particular
        # svn revision via: "svn_revno=commit_sha1", filter them out here:
@@ -786,19 +915,26 @@ sub git_commit {
                git_addremove();
                chomp(my $tree = `git-write-tree`);
                croak if $?;
+               if (exists $tree_map{$tree}) {
+                       my %seen_parent = map { $_ => 1 } @exec_parents;
+                       foreach (@{$tree_map{$tree}}) {
+                               # MAXPARENT is defined to 16 in commit-tree.c:
+                               if ($seen_parent{$_} || @exec_parents > 16) {
+                                       next;
+                               }
+                               push @exec_parents, $_;
+                               $seen_parent{$_} = 1;
+                       }
+               }
                my $msg_fh = IO::File->new_tmpfile or croak $!;
                print $msg_fh $log_msg->{msg}, "\ngit-svn-id: ",
                                        "$SVN_URL\@$log_msg->{revision}",
-                                       " $uuid\n" or croak $!;
+                                       " $SVN_UUID\n" or croak $!;
                $msg_fh->flush == 0 or croak $!;
                seek $msg_fh, 0, 0 or croak $!;
 
-               $ENV{GIT_AUTHOR_NAME} = $ENV{GIT_COMMITTER_NAME} =
-                                               $log_msg->{author};
-               $ENV{GIT_AUTHOR_EMAIL} = $ENV{GIT_COMMITTER_EMAIL} =
-                                               $log_msg->{author}."\@$uuid";
-               $ENV{GIT_AUTHOR_DATE} = $ENV{GIT_COMMITTER_DATE} =
-                                               $log_msg->{date};
+               set_commit_env($log_msg);
+
                my @exec = ('git-commit-tree',$tree);
                push @exec, '-p', $_  foreach @exec_parents;
                open STDIN, '<&', $msg_fh or croak $!;
@@ -814,9 +950,18 @@ sub git_commit {
        if ($commit !~ /^$sha1$/o) {
                croak "Failed to commit, invalid sha1: $commit\n";
        }
-       my @update_ref = ('git-update-ref',"refs/heads/$GIT_SVN-HEAD",$commit);
+       my @update_ref = ('git-update-ref',"refs/remotes/$GIT_SVN",$commit);
        if (my $primary_parent = shift @exec_parents) {
-               push @update_ref, $primary_parent;
+               $pid = fork;
+               defined $pid or croak $!;
+               if (!$pid) {
+                       close STDERR;
+                       close STDOUT;
+                       exec 'git-rev-parse','--verify',
+                                               "refs/remotes/$GIT_SVN^0";
+               }
+               waitpid $pid, 0;
+               push @update_ref, $primary_parent unless $?;
        }
        sys(@update_ref);
        sys('git-update-ref',"$GIT_SVN/revs/$log_msg->{revision}",$commit);
@@ -824,6 +969,16 @@ sub git_commit {
        return $commit;
 }
 
+sub set_commit_env {
+       my ($log_msg) = @_;
+       my $author = $log_msg->{author};
+       my ($name,$email) = defined $users{$author} ?  @{$users{$author}}
+                               : ($author,"$author\@$SVN_UUID");
+       $ENV{GIT_AUTHOR_NAME} = $ENV{GIT_COMMITTER_NAME} = $name;
+       $ENV{GIT_AUTHOR_EMAIL} = $ENV{GIT_COMMITTER_EMAIL} = $email;
+       $ENV{GIT_AUTHOR_DATE} = $ENV{GIT_COMMITTER_DATE} = $log_msg->{date};
+}
+
 sub apply_mod_line_blob {
        my $m = shift;
        if ($m->{mode_b} =~ /^120/) {
@@ -879,21 +1034,106 @@ sub safe_qx {
        return wantarray ? @ret : join('',@ret);
 }
 
-sub svn_check_ignore_externals {
-       return if $_no_ignore_ext;
-       unless (grep /ignore-externals/,(safe_qx(qw(svn co -h)))) {
+sub svn_compat_check {
+       my @co_help = safe_qx(qw(svn co -h));
+       unless (grep /ignore-externals/,@co_help) {
                print STDERR "W: Installed svn version does not support ",
                                "--ignore-externals\n";
                $_no_ignore_ext = 1;
        }
+       if (grep /usage: checkout URL\[\@REV\]/,@co_help) {
+               $_svn_co_url_revs = 1;
+       }
+
+       # I really, really hope nobody hits this...
+       unless (grep /stop-on-copy/, (safe_qx(qw(svn log -h)))) {
+               print STDERR <<'';
+W: The installed svn version does not support the --stop-on-copy flag in
+   the log command.
+   Lets hope the directory you're tracking is not a branch or tag
+   and was never moved within the repository...
+
+               $_no_stop_copy = 1;
+       }
+}
+
+# *sigh*, new versions of svn won't honor -r<rev> without URL@<rev>,
+# (and they won't honor URL@<rev> without -r<rev>, too!)
+sub svn_cmd_checkout {
+       my ($url, $rev, $dir) = @_;
+       my @cmd = ('svn','co', "-r$rev");
+       push @cmd, '--ignore-externals' unless $_no_ignore_ext;
+       $url .= "\@$rev" if $_svn_co_url_revs;
+       sys(@cmd, $url, $dir);
+}
+
+sub check_upgrade_needed {
+       my $old = eval {
+               my $pid = open my $child, '-|';
+               defined $pid or croak $!;
+               if ($pid == 0) {
+                       close STDERR;
+                       exec('git-rev-parse',"$GIT_SVN-HEAD") or croak $?;
+               }
+               my @ret = (<$child>);
+               close $child or croak $?;
+               die $? if $?; # just in case close didn't error out
+               return wantarray ? @ret : join('',@ret);
+       };
+       return unless $old;
+       my $head = eval { safe_qx('git-rev-parse',"refs/remotes/$GIT_SVN") };
+       if ($@ || !$head) {
+               print STDERR "Please run: $0 rebuild --upgrade\n";
+               exit 1;
+       }
 }
+
+# fills %tree_map with a reverse mapping of trees to commits.  Useful
+# for finding parents to commit on.
+sub map_tree_joins {
+       foreach my $br (@_branch_from) {
+               my $pid = open my $pipe, '-|';
+               defined $pid or croak $!;
+               if ($pid == 0) {
+                       exec(qw(git-rev-list --pretty=raw), $br) or croak $?;
+               }
+               while (<$pipe>) {
+                       if (/^commit ($sha1)$/o) {
+                               my $commit = $1;
+                               my ($tree) = (<$pipe> =~ /^tree ($sha1)$/o);
+                               unless (defined $tree) {
+                                       die "Failed to parse commit $commit\n";
+                               }
+                               push @{$tree_map{$tree}}, $commit;
+                       }
+               }
+               close $pipe or croak $?;
+       }
+}
+
+# '<svn username> = real-name <email address>' mapping based on git-svnimport:
+sub load_authors {
+       open my $authors, '<', $_authors or die "Can't open $_authors $!\n";
+       while (<$authors>) {
+               chomp;
+               next unless /^(\S+?)\s*=\s*(.+?)\s*<(.+)>\s*$/;
+               my ($user, $name, $email) = ($1, $2, $3);
+               $users{$user} = [$name, $email];
+       }
+       close $authors or croak $!;
+}
+
 __END__
 
 Data structures:
 
-@svn_log = array of log_msg hashes
+$svn_log hashref (as returned by svn_log_raw)
+{
+       fh => file handle of the log file,
+       state => state of the log file parser (sep/msg/rev/msg_start...)
+}
 
-$log_msg hash
+$log_msg hashref as returned by next_log_entry($svn_log)
 {
        msg => 'whitespace-formatted log entry
 ',                                             # trailing newline is preserved
@@ -911,7 +1151,7 @@ diff-index line ($m hash)
        mode_a => first column of diff-index output, no leading ':',
        mode_b => second column of diff-index output,
        sha1_b => sha1sum of the final blob,
-       chg => change type [MCRAD],
+       chg => change type [MCRADT],
        file_a => original file name of a file (iff chg is 'C' or 'R')
        file_b => new/current file name of a file (any chg)
 }