296963 - Better regression tests for the HTML parser

Status: RESOLVED WORKSFORME

(Core :: DOM: HTML Parser, defect)

Description

[Blake Kaplan (:mrbkap) (inactive)]

While working on bug 272702, I realized that the current regression tests that
we have for the HTML parser are totally useless for any changes that involve
changing the interface between the content sink and the DTD. In other words,
unless your change is restricted solely to the tokenizer and the DTD, the tests
are useless.

I'm filing this bug to make progress on making a new set of regression tests
(maybe pulling in test cases from the layout regression tests, and of course,
the current regression tests) that instead of using the logging sink (which is
why the current tests are inadequate) would serialize the DOM produced by the
content sink.

I know that there is a nsDOMSerializer in extensions/xmlextras, do we want to
use that, or should we just use nsDocumentEncoder directly? Any input welcome.

Comment 1

[Boris Zbarsky [:bzbarsky]]

nsDOMSerializer always serializes as XML.  I suppose that's OK for use -- just
serialize the HTML DOM as XHTML....

Using DOMSerializer is probably simpler than the document encoder and more
future-proof (it's not in need of API cleanup, unlike document encoder).

Comment 2

[Blake Kaplan (:mrbkap) (inactive)]

Mochitests cover this now.