package HTTP::OAI::UserAgent;

use strict;
use warnings;

use vars qw(@ISA $ACCEPT);

# Do not use eval()
our $USE_EVAL = 1;
# Ignore bad utf8 characters
our $IGNORE_BAD_CHARS = 1;
# Silence bad utf8 warnings
our $SILENT_BAD_CHARS = 0;

use constant MAX_UTF8_BYTES => 4;

require LWP::UserAgent;
@ISA = qw(LWP::UserAgent);

unless( $@ ) {
	$ACCEPT = "gzip";
}

sub redirect_ok { 1 }

sub request
{
	my $self = shift;
	my ($request, $arg, $size, $previous, $response) = @_;
	if( ref($request) eq 'HASH' ) {
		$request = HTTP::Request->new(GET => _buildurl(%$request));
	}
	return $self->SUPER::request(@_) unless $response;
	my $parser = XML::LibXML->new(
		Handler => HTTP::OAI::SAXHandler->new(
			Handler => $response->headers
	));
	$parser->{request} = $request;
	$parser->{content_length} = 0;
	$parser->{content_buffer} = Encode::encode('UTF-8','');
	$response->code(200);
	$response->message('lwp_callback');
	$response->headers->set_handler($response);
HTTP::OAI::Debug::trace( $response->verb . " " . ref($parser) . "->parse_chunk()" );
	my $r;
	if( $USE_EVAL ) {
		eval {
			$r = $self->SUPER::request($request,sub {
				$self->lwp_callback( $parser, @_ )
			});
			$self->lwp_endparse( $parser );
		};
	} else {
		$r = $self->SUPER::request($request,sub {
			$self->lwp_callback( $parser, @_ )
		});
		$self->lwp_endparse( $parser );
	}
	if( defined($r) && defined($r->headers->header( 'Client-Aborted' )) && $r->headers->header( 'Client-Aborted' ) eq 'die' )
	{
		$r->code(500);
		$r->message( 'An error occurred while parsing: ' . $r->headers->header( 'X-Died' ));
	}

	$response->headers->set_handler(undef);
	
	# Allow access to the original headers through 'previous'
	$response->previous($r);
	
	my $cnt_len = $parser->{content_length};
	undef $parser;

	# OAI retry-after
	if( defined($r) && $r->code == 503 && defined(my $timeout = $r->headers->header('Retry-After')) ) {
		if( $self->{recursion}++ > 10 ) {
			$self->{recursion} = 0;
			warn ref($self)."::request (retry-after) Given up requesting after 10 retries\n";
			return $response->copy_from( $r );
		}
		if( !$timeout or $timeout =~ /\D/ or $timeout < 0 or $timeout > 86400 ) {
			warn ref($self)." Archive specified an odd duration to wait (\"".($timeout||'null')."\")\n";
			return $response->copy_from( $r );
		}
HTTP::OAI::Debug::trace( "Waiting $timeout seconds" );
		sleep($timeout+10); # We wait an extra 10 secs for safety
		return $self->request($request,undef,undef,undef,$response);
	# Got an empty response
	} elsif( defined($r) && $r->is_success && $cnt_len == 0 ) {
		if( $self->{recursion}++ > 10 ) {
			$self->{recursion} = 0;
			warn ref($self)."::request (empty response) Given up requesting after 10 retries\n";
			return $response->copy_from( $r );
		}
HTTP::OAI::Debug::trace( "Retrying on empty response" );
		sleep(5);
		return $self->request($request,undef,undef,undef,$response);
	# An HTTP error occurred
	} elsif( $r->is_error ) {
		$response->copy_from( $r );
		$response->errors(HTTP::OAI::Error->new(
			code=>$r->code,
			message=>$r->message,
		));
	# An error occurred during parsing
	} elsif( $@ ) {
		$response->code(my $code = $@ =~ /read timeout/ ? 504 : 600);
		$response->message($@);
		$response->errors(HTTP::OAI::Error->new(
			code=>$code,
			message=>$@,
		));
	}

	# Reset the recursion timer
	$self->{recursion} = 0;
	
	# Copy original $request => OAI $response to allow easy
	# access to the requested URL
	$response->request($request);
	$response;
}

sub lwp_badchar
{
	my $codepoint = sprintf('U+%04x', ord($_[2]));
	unless( $SILENT_BAD_CHARS )
	{
		warn "Bad Unicode character $codepoint at byte offset ".$_[1]->{content_length}." from ".$_[1]->{request}->uri."\n";
	}
	return $codepoint;
}

sub lwp_endparse
{
	my( $self, $parser ) = @_; 

	my $utf8 = $parser->{content_buffer};
	# Replace bad chars with '?'
	if( $IGNORE_BAD_CHARS and length($utf8) ) {
		$utf8 = Encode::decode('UTF-8', $utf8, sub { $self->lwp_badchar($parser, @_) });
	}
	if( length($utf8) > 0 )
	{
		_ccchars($utf8); # Fix control chars
		$parser->{content_length} += length($utf8);
		$parser->parse_chunk($utf8);
	}
	delete($parser->{content_buffer});
	$parser->parse_chunk('', 1);
}

sub lwp_callback
{
	my( $self, $parser ) = @_;

	use bytes; # fixing utf-8 will need byte semantics

	$parser->{content_buffer} .= $_[2];

	do
	{
		# FB_QUIET won't split multi-byte chars on input
		my $utf8 = Encode::decode('UTF-8', $parser->{content_buffer}, Encode::FB_QUIET);

		if( length($utf8) > 0 )
		{
			use utf8;
			_ccchars($utf8); # Fix control chars
			$parser->{content_length} += length($utf8);
			$parser->parse_chunk($utf8);
		}

		if( length($parser->{content_buffer}) > MAX_UTF8_BYTES )
		{
			$parser->{content_buffer} =~ s/^([\x80-\xff]{1,4})//s;
			my $badbytes = $1;
			if( length($badbytes) == 0 )
			{
				Carp::confess "Internal error - bad bytes but not in 0x80-0xff range???";
			}
			if( $IGNORE_BAD_CHARS )
			{
				$badbytes = join('', map {
					$self->lwp_badchar($parser, $_)
				} split //, $badbytes);
			}
			$parser->parse_chunk( $badbytes );
		}
	} while( length($parser->{content_buffer}) > MAX_UTF8_BYTES );
}

sub _ccchars {
	$_[0] =~ s/([\x00-\x08\x0b-\x0c\x0e-\x1f])/sprintf("\\%04d",ord($1))/seg;
}

sub _buildurl {
	my %attr = @_;
	Carp::confess "_buildurl requires baseURL" unless $attr{'baseURL'};
	Carp::confess "_buildurl requires verb" unless $attr{'verb'};
	my $uri = new URI(delete($attr{'baseURL'}));
	if( defined($attr{resumptionToken}) && !$attr{force} ) {
		$uri->query_form(verb=>$attr{'verb'},resumptionToken=>$attr{'resumptionToken'});
	} else {
		delete $attr{force};
		# http://www.cshc.ubc.ca/oai/ breaks if verb isn't first, doh
		$uri->query_form(verb=>delete($attr{'verb'}),%attr);
	}
	return $uri->as_string;
}

sub url {
	my $self = shift;
	return _buildurl(@_);
}

sub decompress {
	my ($response) = @_;
	my $type = $response->headers->header("Content-Encoding");
	return $response->{_content_filename} unless defined($type);
	if( $type eq 'gzip' ) {
		my $filename = File::Temp->new( UNLINK => 1 );
		my $gz = Compress::Zlib::gzopen($response->{_content_filename}, "r") or die $!;
		my ($buffer,$c);
		my $fh = IO::File->new($filename,"w");
		binmode($fh,":utf8");
		while( ($c = $gz->gzread($buffer)) > 0 ) {
			print $fh $buffer;
		}
		$fh->close();
		$gz->gzclose();
		die "Error decompressing gziped response: " . $gz->gzerror() if -1 == $c;
		return $response->{_content_filename} = $filename;
	} else {
		die "Unsupported compression returned: $type\n";
	}
}

1;

__END__

=head1 NAME

HTTP::OAI::UserAgent - Extension of the LWP::UserAgent for OAI HTTP requests

=head1 DESCRIPTION

This module provides a simplified mechanism for making requests to an OAI repository, using the existing LWP::UserAgent module.

=head1 SYNOPSIS

	require HTTP::OAI::UserAgent;

	my $ua = new HTTP::OAI::UserAgent;

	my $response = $ua->request(
		baseURL=>'http://arXiv.org/oai1',
		verb=>'ListRecords',
		from=>'2001-08-01',
		until=>'2001-08-31'
	);

	print $response->content;

=head1 METHODS

=over 4

=item $ua = new HTTP::OAI::UserAgent(proxy=>'www-cache',...)

This constructor method returns a new instance of a HTTP::OAI::UserAgent module. All arguments are passed to the L<LWP::UserAgent|LWP::UserAgent> constructor.

=item $r = $ua->request($req)

Requests the HTTP response defined by $req, which is a L<HTTP::Request|HTTP::Request> object.

=item $r = $ua->request(baseURL=>$baseref,verb=>$verb,[from=>$from],[until=>$until],[resumptionToken=>$token],[metadataPrefix=>$mdp],[set=>$set],[oainame=>$oaivalue],...)

Makes an HTTP request to the given OAI server (baseURL) with OAI arguments. Returns an HTTP::Response object.

=item $str = $ua->url(baseURL=>$baseref,verb=>$verb,...)

Takes the same arguments as request, but returns the URL that would be requested.

=back