Package: libhtml-mason-perl
Version: 1:1.35-2
Severity: normal
Tags: patch
Hi,
investigating #387104 and #416097, I have discovered an
incompatibility in the output encoding between HTML::Mason::CGIHandler
and HTML::Mason::ApacheHandler. As far as I understand, CGIHandler is
supposed to emulate ApacheHandler, so I consider this a bug.
The problem is that mod_perl (including mod_perl2) doesn't care about
the Perl internal UTF8 flag on output of non-ASCII strings, but output
coming through HTML::Mason::CGIHandler does. In the CGI case, the output
is converted to ISO-8859-1 automatically, while mod_perl just outputs
the internal encoding of the strings, which happens to be UTF-8.
It can be argued that the mod_perl behaviour is wrong here, but at least
Request Tracker depends on it and thus fails badly when used in a CGI
environment (usually through SpeedyCGI).
I'm attaching the configuration for a minimal testcase that can be
enabled with something like this
# cp masontest.conf /etc/apache/conf.d
# cp handler.cgi cgi.mason /var/www
# ln -s cgi.mason /var/www/mod_perl.mason
# /etc/init.d/apache restart
We now see that the CGI output is in ISO-8859-1 while the mod_perl output
is in UTF-8:
% curl -s http://localhost/cgi.mason | od -c
0000000 344 \n
0000002
% curl -s http://localhost/mod_perl.mason | od -c
0000000 303 244 \n
0000003
The fix is simple: just add "use bytes" into the default output method
of HTML::Mason::CGIHandler. I'm attaching a patch that does this. The
HTML::Mason::CGIHandler documentation indicates that there's no need
for backwards compatibility in this case, but I would be happy with a
way to turn this on optionally if that's preferred.
Please fix this, it's making RT unusable with SpeedyCGI.
Cheers,
--
Niko Tyni [EMAIL PROTECTED]
ScriptAlias /handler.cgi /var/www/handler.cgi
<Location /cgi.mason>
Action html-mason /handler.cgi
SetHandler html-mason
</Location>
PerlModule HTML::Mason::ApacheHandler
<Location /mod_perl.mason>
SetHandler perl-script
PerlHandler HTML::Mason::ApacheHandler
</Location>
<%init>
use charnames ':full';
my $str = "\N{LATIN SMALL LETTER A WITH DIAERESIS}";
</%init>
<% $str %>
#!/usr/bin/perl -w
use strict;
use HTML::Mason::CGIHandler;
my $h = HTML::Mason::CGIHandler->new;
$h->handle_request;
--- /usr/share/perl5/HTML/Mason/CGIHandler.pm 2007/03/29 07:43:25 1.1
+++ /usr/share/perl5/HTML/Mason/CGIHandler.pm 2007/03/29 07:43:46
@@ -99,6 +99,9 @@
$sent_headers = 1;
}
+ # mimic mod_perl behaviour
+ use bytes;
+
# We could perhaps install a new, faster out_method here that
# wouldn't have to keep checking whether headers have been
# sent and what the $r->method is. That would require
