Re: Fwd: [cssom-view] small update

On Mon, 21 Apr 2008 20:33:31 +0200, Garrett Smith <>  
> You've made a spec that disagrees with 3/4 implementations. It sounds
> like you're saying you did it on purpose. Am I right? Please confirm
> that this was deliberate and intentional.

It actually disagrees with all implementations. That's pretty much the  
behavior you get when you define legacy features. See HTML 5 for more  
examples of this.

> Opera doesn't disagree with the spec. You said yourself that you
> changed Opera's implementation then the spec and the spec came first.
> You actually followed up with saying that neither Opera's impl, nor
> CSSOM came first:

Opera does disagree with the specification for a number of features. If I  
remember correctly offset* is not done correctly on inline boxes for  

> [...] The algorithm was backed by an example. Here it is again:
> <body style='position:relative' onload="alert(a.offsetParent ==  
> document.body)">
> <div id=a style='position:relative'>a</div>
> </body>
> The intuitive outcome of the display would be an alert with the value  
> 'true'.
> However, CSSOM would make it so that the outcome would 'false' in the  
> alert.

Actually, per the CSSOM the outcome would be 'true'. Replacing  
a.offsetParent with document.getElementById('a') I get 'true' in Firefox  
and Opera in both quirks and standards mode.

> I've also explained that we need a test suite for these properties and
> why. A test suite provides instant knowledge to everyone. Page
> authors, implementors, and even the spec writer. The test suite, and I
> don't mean one of those "Acid Tests", would be formally composed of a
> test case for each property or set of properties, such as offset*
> would seem to eliminate the need for a spec.

It doesn't eliminate the need for a specification. Boris Zbarsky already  
tried to explain this to you.

I agree that a test suite would be useful. I don't think it should be  
produced before the specification is considered to be done because if the  
specification changes later changing all the tests can be a lot of work.  
The current specification is mostly based on adhoc testing and the tests  
that are publicly available here:

If you think maintaining a test suite against the specification would be  
feasible at this point I encourage you to make one.

> Taking the test approach, there wouldn't be any need for web
> developers to scavenge political discussions on w3c lists for
> artifacts of browser incompatibilities that didn't make it into the
> spec. We (anyone and everyone) could all just load the test in the
> browser and see what's green or red in [browser_x]. The test doesn't
> lie; it would always be up to date with whatever browser it's running
> in.
> I think Anne's answer to that one was "Go start a google code project."

Indeed it was. The CSS WG has no mechanism set up yet apart from to accept tests from contributors and you  
indicated you wanted to use a version control system. So I suggested  
Google Code as I have some experience with that (see html5lib for  

Anne van Kesteren

Received on Monday, 21 April 2008 19:11:21 UTC