[whatwg] text/html flavor conformance checkers and <foo />

[Henri Sivonen]

On Apr 27, 2005, at 04:13, fantasai wrote:

> Henri Sivonen wrote:
>> On Apr 26, 2005, at 19:08, fantasai wrote:
>>> Henri Sivonen wrote:
>>>
>>>> What do you suggest the parser layer of an text/html conformance 
>>>> checker say about <input checkbox ...>?
>>>> 1. Silently treat as <input type="checkbox" ...>?
>>>> 2. Treat as <input type="checkbox" ...> but warn?
>>>> 3. Treat as <input checkbox="checkbox" ...> causing an error to be 
>>>> reported on a higher layer?
>>>> 4. Treat as fatal error in the parser?
>>>> I'm inclined to choose 3.
>>>
>>>
>>> *Why?* Why of all things would you choose to interpret it like 
>>> /that/?
>>> It's neither reporting a useful error, nor handling it per SGML 
>>> rules.
>> To make the separation of concerns similar to what it would be on the 
>> XML side while being real about SGMLness being fiction. That is, the 
>> parser does not need to know if an attribute is allowed. That's a job 
>> for a higher layer.
>
> I still don't understand how this interpretation is useful or required.

It is useful, because it doesn't require knowledge of allowable 
minimizable attributes on the lowest parser level.

> If you want to make <input checkbox> invalid, handle it the same way
> you'd handle <input foo>.

That's what I am suggesting. The parser would treat <input foo> as 
<input foo="foo">, which would be caught on the RELAX NG validation 
layer in my diagram.

> Expanding the attribute from checked to checked="checked" is neither 
> conforming to SGML parsing rules

ITYM checkbox to checkbox="checkbox".

> nor helping the author understand what was wrong.

Would "Attribute 'checkbox' not allowed here." or something along those 
lines be any more incomprehensible that validation errors in general?

> I mean, I understand you're disillusioned with the state of HTML 
> parsing in the world, but it doesn't mean you need to be /reactionary/ 
> about it.

Authors get constantly confused when validator.w3.org feeds them SGML 
fiction. Why shouldn't the QA tools be better aligned with reality?

-- 
Henri Sivonen
hsivonen at iki.fi
http://hsivonen.iki.fi/