[imps] HTML5 parser test location

James Graham jg307 at cam.ac.uk
Mon Apr 7 07:36:53 PDT 2008

Thomas Broyer wrote:
> On Sat, Apr 5, 2008 at 2:17 PM, Anne van Kesteren wrote:
>> Just an FYI since there seems to be some misinformation about this (for
>> which I might have been responsible, not sure), but hereby a note that we'd
>> like to keep all HTML5 parser tests in the html5lib project tree and not
>> move them all to the html5 project tree. For licensing reasons, because the
>> html5 project owner doesn't like it, and because it just isn't worth the
>> trouble.
>> (If I was unclear about this in the past or have given the impression of
>> supporting the opposite view, my apologies.)
> FYI, I started the new tests (not really a "move" 'cause I've been
> rewriting them all from scratch) based on the following thread:
> http://lists.whatwg.org/pipermail/implementors-whatwg.org/2007-July/000126.html
> Particularly the following two messages:
> http://lists.whatwg.org/pipermail/implementors-whatwg.org/2007-July/000127.html
> http://lists.whatwg.org/pipermail/implementors-whatwg.org/2007-August/000129.html

My view on this is:

* The tests must remain MIT licensed. Therefore they can't go in the html5 

* I'm -0 on moving the tests at all. That is, I'm not strongly against it but 
I'm not sure it represents a good investment of time.

* If we do decide to move tests, we should consider the advantages of a 
distributed version control system like Mercurial. It seems like a situation 
where slightly disjoint groups of people are all editing a common set of files 
might play to the strengths of those systems.

* I am totally against rewriting tests. The current tests have often been 
written in response to actual regressions in software. Throwing away all that 
knowledge of fragile points in the various implementations is unacceptable. 
Adding extra tests is of course fine.

* One of the identified problems with the current test suite is that it is hard 
to determine which tests need to change when the spec changes. There are various 
ways to improve this without starting over. Specifically it is not hard to 
instrument html5lib to monitor which phase it is in at a given time. One can 
imagine using this to automatically identify testcases that cause the parser to 
go through altered parts of the spec.

"Eternity's a terrible thought. I mean, where's it all going to end?"
  -- Tom Stoppard, Rosencrantz and Guildenstern are Dead

More information about the Implementors mailing list