W3C home > Mailing lists > Public > public-css-testsuite@w3.org > July 2010

Re: [CSS 2.1] cases that do not pass in any browser

From: Boris Zbarsky <bzbarsky@MIT.EDU>
Date: Wed, 21 Jul 2010 13:38:42 -0400
Message-ID: <4C4730A2.6050507@mit.edu>
To: Arron Eicholz <Arron.Eicholz@microsoft.com>
CC: "public-css-testsuite@w3.org" <public-css-testsuite@w3.org>
On 7/21/10 12:29 PM, Arron Eicholz wrote:

Arron, thanks for looking into these!  Comments inline on the ones I 
know something about (e.g. not the bidi or float tests).

> http://test.csswg.org/suites/css2.1/20100701/html4/abspos-005.htm

This test looks incorrect to me.  The relevant text in section 17.4 is:

   The computed values of properties 'position', 'float', 'margin-*',
   'top', 'right', 'bottom', and 'left' on the table box are used on the
    anonymous box instead of the table box. The table box uses the
   initial values for those properties.

But 'background' is used on the table box, so in this case the 
background doesn't cover the viewport.  I'm not sure how to fix the test.

> http://test.csswg.org/suites/css2.1/20100701/html4/abspos-024.htm

This test is wrong (due to a spec change, I think).  Section 10.3.7 uses 
the direction of the containing block for setting left/right, not the 
direction of the positioned box.  The test should be fixed to reflect 
that.  Of course the test title also needs to be fixed, since it's not 
in fact testing static position in rtl... ;)

> http://test.csswg.org/suites/css2.1/20100701/html4/active-selector-003.htm

This test assumes that script-dispatched DOM events can affect :active 
and the like.  I believe that in Gecko they can't, on purpose.  Can't 
speak to other browsers.

> http://test.csswg.org/suites/css2.1/20100701/html4/at-charset-utf16-be-002.htm
>
> http://test.csswg.org/suites/css2.1/20100701/html4/at-charset-utf16-le-002.htm

These tests should have the red/green in the HTML and linked CSS flipped 
or should just be removed.  See 
https://bugzilla.mozilla.org/show_bug.cgi?id=462458#c6 through 
https://bugzilla.mozilla.org/show_bug.cgi?id=462458#c10

> http://test.csswg.org/suites/css2.1/20100701/html4/first-letter-quote-002.htm

Not passing this is a clear violation of the spec, last I checked.  The 
testcase is correct.

> http://test.csswg.org/suites/css2.1/20100701/html4/first-letter-quote-005.htm

So is this.  This is in fact the same issue as 
first-letter-quote-002.htm, at least in Gecko: the quote and the first 
letter are not part of the same textnode and our first-letter code fails 
to deal.....

> http://test.csswg.org/suites/css2.1/20100701/html4/overflow-applies-to-010.htm

Yeah, I believe no one implements what the spec now says to do for 
'overflow' on list items.  It's the thing that makes "sense" for 
authors, but doesn't necessarily make any sense in the context of the 
rest of CSS and that I thought there was an open issue on (at least in 
terms of marking it at-risk).  The test should stay for now, pending a 
spec change.

> http://test.csswg.org/suites/css2.1/20100701/html4/run-in-breaking-002.htm

This test is correct.  It fails in Gecko because run-in is not 
implemented.  It fails in Webkit because Webkit's run-in implementation 
is completely busted.  It fails in Opera because opera's handling of rtl 
inlines is not really right and becuse it seems to have some additional 
bugs in its run-in implementation specifically.  I haven't looked at it 
in IE for a while.

At what point do we mark run-in as at risk, given the lack of 
interoperable implementations?  ;)

> http://test.csswg.org/suites/css2.1/20100701/html4/run-in-clear-002.htm

This tests reflects what the WG agreed on wrt this behavior, though no 
one's updated the spec in the months since then.  And again, run-in 
behavior in UAs is just broken all over.

I doubt any UA will implement this, but it needs to stay in the test 
suite until the spec is changed or something.

> http://test.csswg.org/suites/css2.1/20100701/html4/tables-width-001.htm

Like this test says, it assumes a sensible table model.  Unfortunately, 
such a model is not web-compatible, so no one implements one.  ;)

> http://test.csswg.org/suites/css2.1/20100701/html4/text-transform-bicameral-004.htm

The only pair that fails this test for me in Gecko, as far as I can see, 
is "I ı", because we lowercase 'I' to 'i' (lowercase latin i) and not to 
lowercase Turkish i.  And since there's no context information that 
would indicate there's Turkish involved anywhere, I believe that 
behavior is correct and the test needs fixing.

> http://test.csswg.org/suites/css2.1/20100701/html4/text-transform-bicameral-005.htm
 > 
http://test.csswg.org/suites/css2.1/20100701/html4/text-transform-bicameral-006.htm

These tests test that given lang="tr" the Turkish rules for uppercasing 
and lowercasing are used.  I believe this is correct per current spec; 
it's just that no UA implements the spec.

> http://test.csswg.org/suites/css2.1/20100701/html4/text-transform-bicameral-021.htm
> http://test.csswg.org/suites/css2.1/20100701/html4/text-transform-bicameral-022.htm

Those look like a lack of Deseret case conversions in UAs.  The test 
looks fine, assuming Deseret is bicameral and Unicode defines the 
case-folding (not my area of expertise).

> http://test.csswg.org/suites/css2.1/20100701/html4/units-008.htm
> http://test.csswg.org/suites/css2.1/20100701/html4/units-009.htm

These tests pass for me in Gecko.

-Boris
Received on Wednesday, 21 July 2010 17:39:16 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 20 September 2010 17:52:02 GMT