miscoded

I've been looking at a bug reported for the Chinese site 56.com (what's up with China and all those numerical domain names - lucky numbers ?) Here's a screenshot from a colleague, Opera on the left and Chrome on the right:


Digging a bit, it boils down to a failure in a piece of JavaScript that reads innerHTML of an element and does a regular expression match() that ends up matching nothing. They use a regular expression created from this string:

var sRepeat="<%begin_"+xLev+"[^>]*%>((.|\\n)+?)<%end_"+xLev+"%>"; 

and it's a problem with how the line endings are encoded. The site's code contains typical Windows line endings (newline + carriage returns) but this regular expression assumes that innerHTML will contain newlines and not carriage returns. (This is the \\n part of the regexp. The fix is simply replacing "\\n" with "\\n|\\r".)

Firefox and Chrome return innerHTML with normalized LF line endings. IE and Opera return CRLF in this case, and comments are broken both in IE (8) and Opera for me.

Wait..did I just say that? We break a shiny, mass-market production site because we follow IE closely on a piece of JavaScript IE invented and defined? And IE is broken too??

Wow! The times they are a-changing..