package CGI::Compress::Gzip; =head1 NAME CGI::Compress::Gzip - CGI with automatically compressed output =head1 LICENSE Copyright 2005 Clotho Advanced Media, Inc., This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself. =head1 SYNOPSIS use CGI::Compress::Gzip; my $cgi = new CGI::Compress::Gzip; print $cgi->header(); print " ..."; =head1 DESCRIPTION [This is beta code, but we use it in our production Linux environments. See the CAVEATS section below for potential gotchas. I've received reports that it fails under Windows. Help!] CGI::Compress::Gzip extends the CGI class to auto-detect whether the client browser wants compressed output and, if so and if the script chooses HTML output, apply gzip compression on any content header for STDOUT. This module is intended to be a drop-in replacement for CGI.pm in a typical scripting environment. Apache mod_perl users may wish to consider the Apache::Compress or Apache::GzipChain modules, which allow more transparent output compression than this module can provide. However, as of this writing those modules are more aggressive about compressing, regardless of Content-Type. =head2 Headers At the time that a header is requested, CGI::Compress::Gzip checks the HTTP_ACCEPT_ENCODING environment variable (passed by Apache). If this variable includes the flag "gzip" and the outgoing mime-type is "text/*", then gzipped output is prefered. [the default mime-type selection of text/* can be changed by subclasses -- see below] The header is altered to add the "Content-Encoding: gzip" flag which indicates that compression is turned on. Naturally, it is crucial that the CGI application output nothing before the header is printed. If this is violated, things will go badly. =head2 Compression When the header is created, this module sets up a new filehandle to accept data. STDOUT is redirected through that filehandle. The new filehandle passes data verbatim until it detects the end of the CGI header. At that time, it switches over to Gzip output for the remainder of the CGI run. Note that the Zlib library on which this code is ultimately based requires a fileno for the output filehandle. Where the output filehandle is faked (i.e. in mod_perl), we instead use in-memory compression. This is more wasteful of RAM, but it is the only solution I've found (and it is one shared by the Apache::* compression modules). Debugging note: if you set B<$CGI::Compress::Gzip::global_give_reason> to a true value, then this module will add an HTTP header entry called B with an explanation of why it chose not to gzip the output stream. =head2 Buffering The Zlib library introduces latencies. In some cases, this module may delay output until the CGI object is garbage collected, presumably at the end of the program. This buffering can be detrimental to long-lived programs which are supposed to have incremental output, causing browser timeouts. To compensate, compression is automatically disabled when autoflush (i.e. the $| variable) is set to true. Future versions may try to enable autoflushing on the Zlib filehandles, if possible [Help wanted]. =cut require 5.005_62; use strict; use warnings; use Carp; use CGI; our @ISA = qw(CGI); our $VERSION = '0.20'; # Package globals our $global_use_compression = 1; # user-settable our $global_can_compress = undef; # 1 = yes, 0 = no, undef = don't know yet # If true, add an outgoing HTTP header explaining why we are not # compressing if gzip turns itself off. our $global_give_reason = 0; #============================== =head1 CLASS METHODS =over 4 =cut #============================== =item new Create a new object. This resets the environment before creating a CGI.pm object. This should not be called more than once per script run! All arguments are passed to the parent class. =cut sub new { my $pkg = shift; $CGI::Compress::Gzip::wrapper::use_fh = undef; select STDOUT; my $self = $pkg->SUPER::new(@_); return $self; } #============================== =item useCompression 1|0 Turn compression on/off for all CGI::Compress::Gzip objects. If turned on, compression will be used only if the prerequisite compression libraries are available and if the client browser requests compression. =cut sub useCompression { my $pkg = shift; my $set = shift; $global_use_compression = $set; return $pkg; } #============================== =back =head1 INSTANCE METHODS =over 4 =cut #============================== =item useFileHandle FILEHANDLE Manually set the output filehandle. Because of limitations of libz, this MUST be a real filehandle (with valid results from fileno()) and not a pseudo filehandle like IO::String. If this is not set, STDOUT is used. =cut sub useFileHandle { my $self = shift; my $fh = shift; $CGI::Compress::Gzip::wrapper::use_fh = $fh; return $self; } #============================== =item isCompressibleType CONTENT-TYPE Given a MIME type (with possible charset attached), return a boolean indicating if this media type is a good candidate for compression. This implementation is simply: return $type =~ /^text\//; Subclasses may wish to override this method to apply different criteria. =cut sub isCompressibleType { my $self = shift; my $type = shift || ""; return $type =~ /^text\//; } #============================== =item header HEADER-ARGS Return a CGI header with the compression flags set properly. Returns an empty string is a header has already been printed. This method engages the Gzip output by fiddling with the default output filehandle. All subsequent output via usual Perl print() will be automatically gzipped except for this header (which must go out as plain text). Any arguments will be passed on to CGI::header. This method should NOT be called if you don't want your header or STDOUT to be fiddled with. =cut sub header { my $self = shift; # further args passed on below if ($self->{'.header_printed'} && $self->{'.zlib_fh'}) { return tied(${$self->{'.zlib_fh'}})->{pending_header}; } my @args = (@_); my $compress = $self->_canCompress(\@args); my $header = $self->SUPER::header(@args); $self->_startCompression($header) if ($compress); return $header; } #============================== # Enable the compression filehandle if: # - The mime-type is appropriate (text/* is the default) # - The programmer wants compression, indicated by the useCompression() # method # - Client wants compression, indicated by the Accepted-Encoding HTTP field # - The IO::Zlib compression library is available sub _canCompress { my $self = shift; my $header = shift; # array ref my $compress = 1; my $reason = ""; my @newheader; # Check programmer preference $compress &&= $global_use_compression; $reason ||= "programmer request" if (!$compress); $CGI::Compress::Gzip::wrapper::flush = $|; # save it in case we change it # Check buffering (disable if autoflushing) $compress &&= (!$CGI::Compress::Gzip::wrapper::flush); $reason ||= "programmer wants unbuffered output" if (!$compress); # Check that browser supports gzip my $acc = $ENV{HTTP_ACCEPT_ENCODING}; $compress &&= ($acc && $acc =~ /\bgzip\b/i); $reason ||= "user agent does not want gzip" if (!$compress); # Check that the output will be HTML $compress &&= $header && ref($header); $reason ||= "no header to check" if (!$compress); my $content_type = ""; if ($compress) { if (@$header && $header->[0] =~ /^[a-z]/) { # Using unkeyed version of arguments - convert to the keyed # version # arg order comes from the header() function in CGI.pm my @flags = qw(Content_Type Status Cookie Target Expires NPH Charset Attachment P3P); for (my $i=0; $i < @$header; $i++) { if ($i < @flags) { push @newheader, "-".$flags[$i], $header->[$i]; } else { # Extra args push @newheader, $header->[$i]; } } } else { @newheader = (@$header); } my $encodingIndex = undef; for (my $i=0; $i < @newheader; $i++) { next if (!defined $newheader[$i]); if ($newheader[$i] =~ /^-?(?:Content[-_]Type|Type)$/i) { $content_type = $newheader[++$i]; } elsif ($newheader[$i] =~ /^-?(?:Content[-_]Type|Type): (.*)$/i) { $content_type = $1; } elsif (($newheader[$i] =~ /^-?Status$/i && $newheader[++$i] =~ /(\d+)/) || $newheader[$i] =~ /^-?Status:\s*(\d+)/i) { my $status = $1; if ($status != 200) { $compress = 0; last; } } elsif (($newheader[$i] =~ /^-?Content[-_]Encoding$/i && ++$i) || $newheader[$i] =~ /^-?Content[-_]Encoding: $/i) { if ($newheader[$i] =~ /\bgzip\b/i) { # Already gzip compressed $compress = 0; last; } else { $encodingIndex = $i; } } } if ($compress) { if (defined $encodingIndex) { $newheader[$encodingIndex] =~ s/^(?:-?Content[-_]Encoding:\s*)/gzip, /mio or $newheader[$encodingIndex] = "gzip, " . $newheader[$encodingIndex]; } else { push @newheader, "-Content_Encoding", "gzip"; } } } $reason ||= "someone already requested gzip" if (!$compress); $content_type ||= "text/html"; if (!$self->isCompressibleType($content_type)) { # Not compressible media $compress = 0; } $reason ||= "incompatible content-type $content_type" if (!$compress); # Check that IO::Zlib is available if ($compress) { if (!defined $global_can_compress) { local $SIG{__WARN__} = 'DEFAULT'; eval "require IO::Zlib"; $global_can_compress = $@ ? 0 : 1; } $compress &&= $global_can_compress; } $reason ||= "IO::Zlib not found" if (!$compress); if ($compress) { @$header = @newheader; } else { push @$header, "-X_non_gzip_reason", $reason if ($global_give_reason); } return $compress; } sub _startCompression { my $self = shift; my $header = shift; $CGI::Compress::Gzip::wrapper::use_fh ||= \*STDOUT; binmode $CGI::Compress::Gzip::wrapper::use_fh; my $filehandle = CGI::Compress::Gzip::wrapper->new($CGI::Compress::Gzip::wrapper::use_fh, "wb"); if (!$filehandle) { carp "Failed to open Zlib output, reverting to uncompressed output"; return undef; } # All output from here on goes to our new filehandle if ($CGI::Compress::Gzip::wrapper::flush && UNIVERSAL::can($filehandle, "autoflush")) { $filehandle->autoflush(); } select $filehandle; $self->{'.zlib_fh'} = $filehandle; # needed for destructor tied(${$self->{'.zlib_fh'}})->{pending_header} = $header; return $self; } #============================== =item DESTROY Override the CGI destructor so we can close the Gzip output stream, if there is one open. =cut sub DESTROY { my $self = shift; if ($self->{'.zlib_fh'}) { $self->{'.zlib_fh'}->close() or &croak("Failed to close the Zlib filehandle"); } return $self->SUPER::DESTROY(); } #============================== package CGI::Compress::Gzip::wrapper; =back =head1 HELPER CLASS CGI::Compress::Gzip also implements a helper class in package CGI::Compress::Gzip::wrapper which subclasses IO::Zlib. This helper is needed to make sure that output is not compressed until the CGI header is emitted. This wrapper delays the ignition of the zlib filter until it sees the exact same header generated by CGI::Compress::Gzip::header() pass through it's WRITE() method. If you change the header before printing it, this class will throw an exception. This class hold one global variable representing the previous default filehandle used before the gzip filter is put in place. This filehandle, usually STDOUT, is replaced after the gzip stream finishes (which is usually when the CGI object goes out of scope and is destroyed). =cut our @ISA = qw(IO::Zlib); our $use_fh; sub OPEN { my $self = shift; # Delay opening until after the header is printed. $self->{openargs} = [@_]; $self->{outtype} = undef; $self->{buffer} = ""; return $self; } sub WRITE { my $self = shift; my $buf = shift; my $length = shift; my $offset = shift; # Appropriated from IO::Zlib: &Carp::croak("bad LENGTH") unless ($length <= length($buf)); &Carp::croak("OFFSET not supported") if (defined($offset) && $offset != 0); my $bytes = 0; my $header = $self->{pending_header}; if ($header) { if (length($header) > $length) { $self->{pending_header} = substr($header, $length); $header = substr($header, 0, $length); } else { $self->{pending_header} = ""; } if ($buf =~ s/^\Q$header//s) { no strict qw(refs); if (print $use_fh $header) { $bytes += length($header); $length -= length($header); } else { &Carp::croak("Failed to print the uncompressed CGI header"); } } else { &Carp::croak("Expected to print the CGI header"); } } if ($length) { if (!defined $self->{outtype}) { my $fh = $self->{openargs}->[0]; my $mod_perl = ($ENV{MOD_PERL} || ($ENV{GATEWAY_INTERFACE} && $ENV{GATEWAY_INTERFACE} =~ /^CGI-Perl\//)); my $isglob = (ref($fh) && ref($fh) eq "GLOB" && defined $fh->fileno()); my $isfilehandle = (ref($fh) && ref($fh) !~ /^GLOB|SCALAR|HASH|ARRAY|CODE$/ && $fh->can("fileno") && defined $fh->fileno()); if ((!$mod_perl) && ($isglob || $isfilehandle)) { # Finished printing header! # Complete delayed open if (!$self->SUPER::OPEN(@{$self->{openargs}})) { &Carp::croak("Failed to open the compressed output stream"); } $self->{outtype} = "stream"; } else { $self->{outtype} = "block"; } } if ($self->{outtype} eq "stream") { $bytes += $self->SUPER::WRITE($buf, $length, $offset); } else { $self->{buffer} .= $buf; $bytes += length $buf; } } return $bytes; } sub CLOSE { # Debugging: #$SIG{__DIE__} = sub {print STDERR "die: ".join("", @_)}; my $self = shift; $use_fh = undef; select STDOUT; my $result; if ($self->{outtype} && $self->{outtype} eq "stream") { $result = $self->SUPER::CLOSE(); if (!$result) { &Carp::confess("Failed to close gzip $!"); } } else { print STDOUT Compress::Zlib::memGzip($self->{buffer}); $result = 1; } return $result; } 1; __END__ =head1 CAVEATS =head2 Windows This module fails tests specifically under Windows in ways I do not understand, as reported by CPANPLUS testers. There is some problem with my IO::Zlib tests. If anyone knows about IO::Zlib failures or caveats on Windows, please let me know. It *might* be related to binmode, but I have not tested this theory. =head2 Apache::Registry Under Apache::Registry, global variables may not go out of scope in time. This may causes timing bugs, since this module makes use of the DESTROY() method. To avoid this issue, make sure your CGI object is stored in a scoped variable. # BROKEN CODE use CGI::Compress::Gzip; $q = CGI::Compress::Gzip->new; print $q->header; print "Hello, world\n"; # WORKAROUND CODE use CGI::Compress::Gzip; do { my $q = CGI::Compress::Gzip->new; print $q->header; print "Hello, world\n"; } =head2 Filehandles This module works by changing the default filehandle. It does not change STDOUT at all. As a consequence, your programs should call C without a filehandle argument. # BROKEN CODE use CGI::Compress::Gzip; my $q = CGI::Compress::Gzip->new; print STDOUT $q->header; print STDOUT "Hello, world\n"; # WORKAROUND CODE use CGI::Compress::Gzip; my $q = CGI::Compress::Gzip->new; print $q->header; print "Hello, world\n"; Future versions may steal away STDOUT and replace it with the compression filehandle, but that seemed too risky for this version. =head1 SEE ALSO CGI::Compress::Gzip depends on CGI and IO::Zlib. Similar functionality is available from mod_gzip, Apache::Compress or Apache::GzipChain, however all of those require changes to the webserver's configuration. =head1 AUTHOR Clotho Advanced Media, I Primary developer: Chris Dolan =head1 THANKS Clotho greatly appeciates the assistance and feedback the community has extended to help refine this module. Thanks to Rhesa Rozendaal who noticed the -Type omission in v0.17. Thanks to Laga Mahesa who did some Windows testing and expirimentation. Thanks to Slaven Rezic who 1) found several header handling bugs, 2) discovered the Apache::Registry and Filehandle caveats, and 3) provided a patch incorporated into v0.17. Thanks to Jan Willamowius who found a header handling bug. Thanks to Andreas J. Koenig and brian d foy for module naming advice. =head1 HELP WANTED If you like this module, please help by testing on Windows or in a FastCGI environment, since I have neither available for easy testing.