#!@PERL@
#
# checklinks.pl
#
# This script extracts all links from a HTML page and checks their validity.
# Written to use 'curl' for URL checking.
#
# Author: Daniel Stenberg <Daniel.Stenberg@sth.frontec.se>
# Version: 0.7 Sept 30, 1998
#
# HISTORY
#
# 0.5 - Cuts off the #-part from links before checking.
#
# 0.6 - Now deals with error codes 3XX better and follows the Location:
#       properly.
#     - Added the -x flag that only checks http:// -links
#
# 0.7 - Ok, http://www.viunga.se/main.html didn't realize this had no path
#       but a document. Now it does.
#
#

$in="";

 argv:
if($ARGV[0] eq "-v" ) {
    $verbose = 1;
    shift @ARGV;
    goto argv;
}
elsif($ARGV[0] eq "-i" ) {
    $usestdin = 1;
    shift @ARGV;
    goto argv;
}
elsif($ARGV[0] eq "-l" ) {
    $linenumber = 1;
    shift @ARGV;
    goto argv;
}
elsif($ARGV[0] eq "-h" ) {
    $help = 1;
    shift @ARGV;
    goto argv;
}
elsif($ARGV[0] eq "-x" ) {
    $external = 1;
    shift @ARGV;
    goto argv;
}

$geturl = $ARGV[0];

if(($geturl eq "") || $help) {
    print  "Usage: $0 [-hilvx] <full URL>\n",
    " Use a traling slash for directory URLs!\n",
    " -h  This help text\n",
    " -i  Read the initial page from stdin\n",
    " -l  Line number report for BAD links\n",
    " -v  Verbose mode\n",
    " -x  Check non-local (external?) links only\n";
    exit;
}

if($ARGV[1] eq "-") {
    print "We use stdin!\n";
    $usestdin = 1;
}

# This is necessary from where I tried this:
#$proxy =" -x 194.237.142.41:80";

# linkchecker, URL will be appended to the right of this command line
# this is the one using HEAD:
$linkcheck = "curl -s -m 20 -I$proxy";

# as a second attempt, this will be used. This is not using HEAD but will
# get the whole frigging document!
$linkcheckfull = "curl -s -m 20 -i$proxy";

# htmlget, URL will be appended to the right of this command line
$htmlget = "curl -s$proxy";

# Parse the input URL and split it into the relevant parts:

sub SplitURL {
    my $inurl = $_[0];

    if($inurl=~ /^([^:]+):\/\/([^\/]*)\/(.*)\/(.*)/ ) {
        $getprotocol = $1;
        $getserver = $2;
        $getpath = $3;
        $getdocument = $4;
    }
    elsif ($inurl=~ /^([^:]+):\/\/([^\/]*)\/(.*)/ ) {
        $getprotocol = $1;
        $getserver = $2;
        $getpath = $3;
        $getdocument = "";

        if($getpath !~ /\//) {
            $getpath ="";
            $getdocument = $3;
        }

    }
    elsif ($inurl=~ /^([^:]+):\/\/(.*)/ ) {
        $getprotocol = $1;
        $getserver = $2;
        $getpath = "";
        $getdocument = "";
    }
    else {
        print "Couldn't parse the specified URL, retry please!\n";
        exit;
    }
}

&SplitURL($geturl);

#print "protocol = $getprotocol\n";
#print "server = $getserver\n";
#print "path = $getpath\n";
#print "document = $getdocument\n";
#exit;

if(!$usestdin) {
    open(HEADGET, "$linkcheck $geturl|") ||
        die "Couldn't get web page for some reason";
  headget:
    while(<HEADGET>) {
#       print $_;
        if($_ =~ /HTTP\/.*3\d\d /) {
            $pagemoved=1;
        }
        elsif($pagemoved &&
            ($_ =~ /^Location: (.*)/)) {
            $geturl = $1;

            &SplitURL($geturl);

            $pagemoved++;
            last headget;
        }
    }
    close(HEADGET);

    if($pagemoved == 1) {
        print "Page is moved but we don't know where. Did you forget the ",
            "traling slash?\n";
        exit;
    }

    open(WEBGET, "$htmlget $geturl|") ||
        die "Couldn't get web page for some reason";

    while(<WEBGET>) {
        $line = $_;
        push @indoc, $line;
        $line=~ s/\n//g;
        $line=~ s/\r//g;
#       print $line."\n";
        $in=$in.$line;
    }

    close(WEBGET);
}
else {
    while(<STDIN>) {
        $line = $_;
        push @indoc, $line;
        $line=~ s/\n//g;
        $line=~ s/\r//g;
        $in=$in.$line;
    }
}

#print length($in)."\n";

sub LinkWorks {
    my $check = $_[0];

#   URL encode:
#    $check =~s/([^a-zA-Z0-9_:\/.-])/uc sprintf("%%%02x",ord($1))/eg;

    @doc = `$linkcheck \"$check\"`;

    $head = 1;

#    print "COMMAND: $linkcheck \"$check\"\n";
#    print $doc[0]."\n";

  boo:
    if( $doc[0] =~ /^HTTP[^ ]+ (\d+)/ ) {
        $error = $1;

        if($error < 400 ) {
            return "GOOD";
        }
        else {

            if($head && ($error >= 500)) {
                # This server doesn't like HEAD!
                @doc = `$linkcheckfull \"$check\"`;
                $head = 0;
                goto boo;
            }
            return "BAD";
        }
    }
    return "BAD";
}


sub GetLinks {
    my $in = $_[0];
    my @result;

  getlinkloop:
    while($in =~ /[^<]*(<[^>]+>)/g ) {
        # we have a tag in $1
        $tag = $1;
    
        if($tag =~ /^<!--/) {
            # this is a comment tag, ignore it
        }
        else {
            if($tag =~ /(src|href|background|archive) *= *(\"[^\"]\"|[^ )>]*)/i) {
                $url=$2;
                if($url =~ /^\"(.*)\"$/) {
                    # this was a "string" now $1 has removed the quotes:
                    $url=$1;
                }


                $url =~ s/([^\#]*)\#.*/$1/g;

                if($url eq "") {
                    # if the link was nothing than a #-link it may now have
                    # been emptied completely so then we skip the rest
                    next getlinkloop;
                }

                if($done{$url}) {
                    # if this url already is done, do next
                    $done{$url}++;
                    next getlinkloop;
                }

                $done{$url} = 1; # this is "done"

                push @result, $url;
                if($tag =~ /< *([^ ]+)/) {
#                   print "TAG: $1\n";
                    $tagtype{$url}=$1;
                }
            }
        }
    }
    return @result;
}

@links = &GetLinks($in);

 linkloop:
for(@links) {
    $url = $_;

    if($url =~ /^([^:]+):/) {
        $prot = $1;
#       if($prot !~ /(http|ftp)/i) {
        if($prot !~ /http/i) {
            # this is an unsupported protocol, we ignore this
            next linkloop;
        }
        $link = $url;
    }
    else {
        if($external) {
            next linkloop;
        }

        # this is a link on the save server:
        if($url =~ /^\//) {
            # from root
            $link = "$getprotocol://$getserver$url";
        }
        else {
            # from the scanned page's dir
            $nyurl=$url;

            if(length($getpath) &&
               ($getpath !~ /\/$/) &&
               ($nyurl !~ /^\//)) {
                # lacks ending slash, add one to the document part:
                $nyurl = "/".$nyurl;
            }
            $link = "$getprotocol://$getserver/$getpath$nyurl";
        }
    }

#print "test $link\n";
#$success = "GOOD";

    $success = &LinkWorks($link);

    $count = $done{$url};

    $allcount += $count;

    print "$success $count <".$tagtype{$url}."> $link $url\n";

# If bad and -l, present the line numbers of the usage
    if("BAD" eq $success) {
        $badlinks++;
        if($linenumber) {
            $line =1;
            for(@indoc) {
                if($_ =~ /$url/) {
                    print " line $line\n";
                }
                $line++;
            }
        }
    }

}

if($verbose) {
    print "$allcount links were checked";
    if($badlinks > 0) {
        print ", $badlinks were found bad";
    }
    print "\n";
}