On Tue, Dec 04, 2007 at 08:16:24AM +1030, Benjamin Close wrote:
Jakub Narebski wrote:
On Mon, 3 Dec 2007, Martin Koegler wrote:
On Mon, Dec 03, 2007 at 04:06:48AM -0800, Jakub Narebski wrote:
Ismail Dönmez <ismail(_at_)pardus(_dot_)org(_dot_)tr> writes:
Monday 03 December 2007 Tarihinde 12:14:43 yazm??t?:
Benjamin Close <Benjamin(_dot_)Close(_at_)clearchain(_dot_)com> writes:
- eval { $res = decode_utf8($str, Encode::FB_CROAK); };
- if (defined $res) {
- return $res;
- } else {
- return decode($fallback_encoding, $str,
Encode::FB_DEFAULT);
- }
+ eval { return ($res = decode_utf8($str, Encode::FB_CROAK));
};
+ return decode($fallback_encoding, $str, Encode::FB_DEFAULT);
}
This version is broken on Debian sarge and etch. Feeding a UTF-8 and a
latin1
encoding of the same character sequence yields to different results.
For the record, this was on a debian sid machine.
#perl --version
This is perl, v5.8.8 built for x86_64-linux-gnu-thread-multi
and the result of not using the original patch was:
<h1>Software error:</h1>
<pre>Cannot decode string with wide characters at
/usr/lib/perl/5.8/Encode.pm line 166.
</pre>
I haven't tried the other solutions tested here.
Debian etch also has v5.8.8.
My main question is, why is the error not catched?
I'm not a perl programmer, but in your patch the first line is a
NOP. The return in eval seems to only returns from the eval block, so
any text is decoded as latin1 with the second statement.
In the original version, decode($fallback_encoding, $str,
Encode::FB_DEFAULT) can not emit an error, else it would in your
version too.
In your version, eval is able to surpress the error of
decode_utf8($str, Encode::FB_CROAK);, but not in the original version.
Strange.
mfg Martin Kögler
