$a = "foo\x{100}";
chop($a);
print " : ", md5_hex("");
print "$a: ", md5_hex($a);
print "foo: ", md5_hex("foo");
__END__
Results in:
Digest::MD5 changed to use SvPVbyte to get the value of the passed SV, instead of SvPV, and it totally broke in perl 5.6. For some reason, the fact that $a has UTF8 flag on makes it blow up. Fun fun fun! Gisle's looking into it; at the very least, maybe it can be changed to use SvPV for < 5.7, but still.
The original test case was using the result of XML::RSS / XML::Parser. Should XML::Parser be flagging SVs UTF8 in the first place, especially if there's no high-bit characters in them?
