X-Git-Url: https://git.madduck.net/code/myrepos.git/blobdiff_plain/af9a9d12caffaccb04e0e989a5b71e18370953f4..3d39c568b0f5559c4bf1e3c054588ca255c26e21:/webcheckout?ds=sidebyside diff --git a/webcheckout b/webcheckout index 33bccb9..0c93eef 100755 --- a/webcheckout +++ b/webcheckout @@ -1,19 +1,89 @@ #!/usr/bin/perl + +=head1 NAME + +webcheckout - check out repositories referenced on a web page + +=head1 SYNOPSIS + +B [options] url [destdir] + +=head1 DESCRIPTION + +B downloads an url and parses it, looking for version control +repositories referenced by the page. It checks out each repository into +a subdirectory of the current directory, using whatever VCS program is +appropriate for that repository (git, svn, etc). + +The information about the repositories is embedded in the web page using +the rel=vcs-* microformat, which is documented at +. + +If the optional destdir parameter is specified, VCS programs will be asked +to check out repositories into that directory. If there are multiple +repositories to check out, each will be checked out into a separate +subdirectory of the destdir. + +=head1 OPTIONS + +=over 4 + +=item -a, --auth + +Prefer authenticated repositories. By default, webcheckout will use +anonymous repositories when possible. If you have an account that +allows you to use authenticated repositories, you might want to use this +option. + +=item --no-act, -n + +Do not actually check anything out, just print out the commands that would +be run to check out the repositories. + +=item --quiet, -q + +Quiet mode. Do not print out the commands being run. (The VCS commands +may still be noisy however.) + +=back + +=head1 PREREQUISITES + +To use this program you will need lots of VCS programs installed, +obviously. It also depends on the perl LWP and HTML::Parser modules. + +If the perl URI module is installed, webcheckout can heuristically guess +what you mean by partial URLs, such as "kitenet.net/~joey"' + +=head1 AUTHOR + +Copyright 2009 Joey Hess + +Licensed under the GNU GPL version 2 or higher. + +This program is included in mr + +=cut + use LWP::Simple; use HTML::Parser; +use Getopt::Long; use warnings; use strict; +# What to download. +my $url; + # Controls whether to print what is being done. -my $verbose=1; +my $quiet=0; # Controls whether to actually check anything out. -my $noact=1; +my $noact=0; # Controls whether to perfer repos that use authentication. my $want_auth=0; -# Controls where to check out to. If not set, the vcs is allowed to +# Controls where to check out to. If not set, the VCS is allowed to # decide. my $destdir; @@ -35,9 +105,33 @@ my @anon_urls=( qr/^http:\/\//i, # generally the worst transport ); +sub getopts { + Getopt::Long::Configure("bundling", "no_permute"); + my $result=GetOptions( + "q|quiet" => \$quiet, + "n|noact" => \$noact, + "a|auth", => \$want_auth, + ); + if (! $result || @ARGV < 1) { + die "usage: webcheckout [options] url [destdir]\n"; + } + + $url=shift @ARGV; + $destdir=shift @ARGV; + + eval q{use URI::Heuristic}; + if (! $@) { + $url=URI::Heuristic::uf_uristr($url); + } + + if ($noact) { + $quiet=0; + } +} + sub doit { my @args=grep { defined } @_; - print join(" ", @args)."\n" if $verbose; + print join(" ", @args)."\n" unless $quiet; return 0 if $noact; return system(@args); } @@ -46,23 +140,26 @@ sub doit { sub better { my ($a, $b)=@_; - my $firstanon=$b; + my @anon; foreach my $r (@anon_urls) { if ($a->{href} =~ /$r/) { - $firstanon=$a; - last; + push @anon, $a; } elsif ($b->{href} =~ /$r/) { - $firstanon=$b; - last; + push @anon, $b; } } if ($want_auth) { - return $firstanon != $a; + # Whichever is authed is better. + return 1 if ! @anon || ! grep { $_ eq $a } @anon; + return 0 if ! grep { $_ eq $b } @anon; + # Neither is authed, so the better anon method wins. + return $anon[0] == $a; } else { - return $firstanon == $a; + # Better anon method wins. + return @anon && $anon[0] == $a; } } @@ -71,21 +168,26 @@ sub better { sub dedup { my %seenhref; my %bytitle; + my @others; foreach my $repo (@_) { if (exists $repo->{title} && - length $repo->{title} && - exists $bytitle{$repo->{title}}) { - my $other=$bytitle{$repo->{title}}; - next unless better($repo, $other); - delete $bytitle{$other->{title}} - } + length $repo->{title}) { + if (exists $bytitle{$repo->{title}}) { + my $other=$bytitle{$repo->{title}}; + next unless better($repo, $other); + delete $bytitle{$other->{title}} + } - if (! $seenhref{$repo->{href}}++) { - $bytitle{$repo->{title}}=$repo; + if (! $seenhref{$repo->{href}}++) { + $bytitle{$repo->{title}}=$repo; + } + } + else { + push @others, $repo; } } - return values %bytitle; + return values %bytitle, @others; } sub parse { @@ -93,25 +195,45 @@ sub parse { my @ret; my $parser=HTML::Parser->new(api_version => 3); + my $abody=undef; + my $aref=undef; $parser->handler(start => sub { my $tagname=shift; my $attr=shift; - return if lc $tagname ne 'link'; - return if ! exists $attr->{rel} || lc $attr->{rel} ne 'vcs'; + return if ! exists $attr->{href} || ! length $attr->{href}; - return if ! exists $attr->{type} || ! length $attr->{type}; + return if ! exists $attr->{rel} || $attr->{rel} !~ /^vcs-(.+)/i; + $attr->{type}=lc($1); + + # need to collect the body of the tag if there is no title + if ($tagname eq "a" && ! exists $attr->{title}) { + $abody=""; + $aref=$attr; + } + push @ret, $attr; }, "tagname, attr"); + $parser->handler(text => sub { + if (defined $aref) { + $abody.=join(" ", @_); + } + }, "text"); + $parser->handler(end => sub { + my $tagname=shift; + if ($tagname eq "a" && defined $aref) { + $aref->{title}=$abody; + $aref=undef; + $abody=undef; + } + }, "tagname"); + $parser->report_tags(qw{link a}); $parser->parse($page); $parser->eof; return @ret; } -my $url=shift; -if (! defined $url) { - die "usage: webcheckout url\n"; -} +getopts(); my $page=get($url); if (! defined $page) { @@ -123,6 +245,19 @@ if (! @repos) { die "no repositories found on $url\n"; } +#use Data::Dumper; +#print Dumper(\@repos); +#exit; + +if (defined $destdir && @repos > 1) { + # create subdirs of $destdir for the multiple repos + if (! $noact) { + mkdir($destdir); + chdir($destdir) || die "failed to chdir to $destdir: $!"; + } + $destdir=undef; +} + my $errors=0; foreach my $repo (@repos) { my $handler=$handlers{$repo->{type}}; @@ -139,6 +274,3 @@ foreach my $repo (@repos) { } } exit($errors > 0); - -#use Data::Dumper; -#print Dumper(\@repos);