Re: ISSUE-54: doctype-legacy-compat

Robin Berjon wrote:

 > As I understand it, your objection is entirely built on top of the notion that it is better for a format to carry a version identifier, or in fact even more strongly that it is wrong for a format not to carry a version identifier.

I'm not sure about "a format" but I think that is
just an issue of terminology.  In words that I understand,
I think it wrong for a document that purports to be
HTML to fail to assert to which version of HTML
it claims adherance.  I don't want to generalise
this to documents in general, because we have an
excellent precedent for HTML but far less (or
no) precedent for (say) FORTRAN programs.

 > You cite the example of validation, which I doubt applies. One always validates against something specific — you're not going to validate HTML against SVG or PNG — and that specificity includes the version. What does adding a version identifier bring to the table here? If I'm your resident QA nazi and I say that you have to validate to v5, I'm not going to care that your document labelled itself v8.

A document that was valid when it was written should
not cease to be valid if the specification changes
under its feet.

 > The case for a version identifier that would be used by the user agent is even more tenuous. If the document labels itself as v5 but contains v6 content, should a UA that understands v6 skip the newer features? I think that's of dubious value, and something tells me that the browser vendors might scream a little bit.

Much as I would like UAs to adjust their behaviour based on
the DOCTYPE (and we could avoid many tedious discussions in
this if implementors were willing to make this the case),
I am not going to hold my breath waiting for them so to do.

But by /providing/ version information to a UA, we /allow/
it to modify its behaviour; if we fail to provide that information,
then we cannot complain if it parses the document as per the
version of the specification that was current when it (the
UA) was written.

 > Furthermore, experience in the wild (notably with SVG) shows that as soon as you have two versions a non-negligible subset of all documents start being labelled with the wrong version, meaning you now have a lot of useless metadata on your hands.

This can happen anyway; documents that were written
to HTML 4.01 (or earlier) could be re-labelled by
well-meaning document owners as <!DOCTYPE html> just
as easily as an HTML 5.0 document could be mis-labelled
HTML 5.01.  As soon as we allow authors to type a single
byte we allow them to make errors : I do not see why
we should penalise careful document authors in order
to make it harder for careless ones to make mistakes ...

 > Version identifiers don't really help for this sort of context, they're just extra bytes.

I understand your perspective; I simply don't share it !

Philip TAYLOR

Received on Tuesday, 20 January 2009 12:29:40 UTC