[subbu@earth lib] echo "equation C<r in <ref>foo</ref>" | node parse --normalize <p>equation C<r in <ref>foo</ref></p>
Causes the diff: https://fr.wikipedia.org/w/index.php?title=%C3%89volution_de_l%27altruisme&diff=prev&oldid=112044063 as reported on the WP:VE/F page.
This is parsing as <r in="" <ref="">foo</ref> and then the r tag is being sanitized.
We changed to this behaviour in https://gerrit.wikimedia.org/r/#/c/173212/. See the output for "Handle broken pre-like tags (bug 64025)".
The parsing spec (http://www.w3.org/TR/html5/syntax.html#attributes-0) doesn't omit < from attribute names.
However, the php parser clearly differs here:
<div id="1" <div id="2">ine</div>
becomes
<div id="1" <div id="2"> ine </div>
So, we could return to breaking on < or try and "fix" this on the php side. Any thoughts?
I verified in Firefox that it parses the div just like we do .. so, yes, the PHP parser is one that parses this differently. I wonder if it is because of HTML4 .. But, if we change the PHP parser behavior (which seems reasonable), that could break a lot of exisitng pages. So, not sure yet ..
Perhaps a grep of dumps can help inform whether we change PHP parser behavior or fix Parsoid to deviate from HTML5 syntax.