Re: conformance (was layout solutions blah blah blah)

Tantek Çelik responded to me:

>> Conformance criteria are clear in CSS1 section 7.  There is
>> no mention in that section that the test suite is the measure
>> of conformance.  There is no correction or
>> supplementary Recommendation that defines the test suite as
>> the measure of conformance.
> 
> That's why I said _objective_.  Following
> and evaluating prose is subjective, and tends
> to be more difficult to prove/disprove.

I take the CSS1 test suite as a measure of *non-conformance*.  
That is, each implementation that fails to pass the test suite is 
non-conformant.  The corollary is that each 
conformant implementation passes the test suite.  It is 
not necessarily the case that an implementation passing the 
test suite is conformant.

The World Wide Web Consortium is not in the business 
of monitoring or certifying conformance to CSS.  Indeed, there is 
no organization in that business.  So, test suite or no test suite, it 
all comes down to the claims of vendors and the evaluations 
by discerning individuals of those claims.  I grant that having 
a centralized and well-known test suite reduces the probability 
of disagreement over conformance, but I stress that in every case 
it is the individual who passes judgment.

> Though, in principle, I agree somewhat in that conformance
> is noted by the spec, but without testable instances of that
> conformance it is of significantly less value.

I do not doubt the value of a conformance test, but I believe that 
a comprehensive test will be extraordinarily difficult to design 
for CSS1, to say nothing of CSS2 and later versions.  When 
we multiply the methods of incorporating CSS times the units 
of functionality times the lexical variations times the element 
types, we get tens of thousands of cases.  Who would traverse 
the entirety of such a test suite?  Then we must check 
error handling.  The number of possible errors is infinite.  No 
person can create or traverse an infinite test suite.  So what are 
we to do?

>> I do not want to clutter www-style with
>> the details of implementation bugs.
> 
> Actually, when it comes to specs which have yet to exit CR
> (see my previous comment about such specs),
> this is exactly the place to discuss it.

Yes, now that I consider it from your perspective, that is 
very sensible.

> If implementations are having difficulty or are unable to
> implement a particular feature, it probably means it should
> be removed from the current version of the spec and put back
> on the drawing board for the next version. This is not just
> an academic exercise afterall.

So are we to release CSS1.1 and CSS2.1?  I mean, I quite agree 
with your opinion as concerns specifications officially 
under development.  But how do we retroactively apply those 
same criteria to specifications released in what was considered 
(and by many still is considered) final form?

>> (Confer "The Rise of 'Worse is Better' ''
>> <http://www.ai.mit.edu/docs/articles/good-new
s/subsection3.2.1.html>.)
> 
> Funny article.  Seems like it can be summarized by 
short statement:
> 
> You can either do something reasonable and then improve
> it over time, or sit around all day thinking about how
> to do it perfectly and never actually do it.
> 
> Evolution predicts the former will succeed.  This is nothing new.

I like your summary, but mine would be:

Social factors are as important as technical factors in 
determining the success of software.

This is nothing new, either, but Richard P. Gabriel makes 
an eloquent case.  He recognizes that, even when people 
finish thinking about a technical marvel and implement it, it 
can suffer from neglect and die.

-- 
Etan Wexler

Received on Thursday, 7 February 2002 05:54:10 UTC