Re: Formal definition of HTML5 (was Re: Version information)

On Tue, 17 Apr 2007, Henrik Dvergsdal wrote:
> As I indicated earlier: There will always be aspects of programming 
> languages (and programs) that aren't automatically checked.

Yes, but we want these aspects to be as few and far between as possible. 
We want to encourage an ecosystem where conformance checkers compete over 
how many errors they can test for, in the same way that browsers compete 
in how many test cases they pass.

> This is just a question of where to draw the line.

And in my opinion, we, as a working group, shouldn't draw that line.

> I cannot see how having an official schema (or an official set of 
> schemas) imposes any limits on our tools.

HTML4 has an official schema, and in the YEARS of HTML4 being the latest 
and greatest, no serious usable conformance checker was made that used 
more than the official schema. HTML5 isn't even complete yet and we 
ALREADY have a conformance checker that goes well beyond what the HTML4 
validator tests.

Thus, historical evidence suggests that having an official schema imposes 
a limit, even if it is purely a psychological one.

> > (And if we do have a spec schema, and it doesn't catch everything, you 
> > know people will claim that conformance checkers that catch mistakes 
> > the spec schema wouldn't flag are buggy and are reporting bogus 
> > errors.
> I don't think so.

People DO claim this about HTML4. I have repeatedly run into people who, 
when informed that their Web pages were non-conformant, said "but it 
validates!". When told that the validator was buggy, they usually dismiss 
such claims by saying "the validator uses the DTD", and by asserting that 
the DTD is the be all and end all of conformance checking.

By avoiding having an official schema at all, we short circuit that entire 
line of reasoning. The net result will, in my opinion, be a marked 
improvement in the quality of conformance checking and thus the quality of 
Web pages as a whole.

> If they wonder about errors from a conformance checker, they will 
> consult the spec and not the schema, especially if the errors are 
> accompanied by text encouraging them to do so.

In practice, they do not wonder about errors. The problem is that the 
conformance checkers using spec-provided schemas fail to catch errors, and 
it is those errors that they deny.

Ian Hickson               U+1047E                )\._.,--....,'``.    fL       U+263A                /,   _.. \   _\  ;`._ ,.
Things that are impossible just take longer.   `._.-(,_..'--(,_..'`-.;.'

Received on Tuesday, 17 April 2007 08:05:05 UTC