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

RE: FW: In a quick glance, all of the whitespace test file are invalid tests that don't prove anything

From: Gérard Talbot <css21testsuite@gtalbot.org>
Date: Tue, 2 Mar 2010 10:50:34 -0800
Message-ID: <041d7591bd1cec4c9e1f5724ff5852c6.squirrel@cp3.shieldhost.com>
To: "Arron Eicholz" <Arron.Eicholz@microsoft.com>
Cc: "public-css-testsuite@w3.org" <public-css-testsuite@w3.org>
Hello all,

>> Gérard Talbot wrote:

>> This is the same problem with testcases which expect the tester to be
>> able to
>> see, measure (or compare) a 9px (or 12px or 18px) difference between 2
>> objects with the naked eye.

> Arron Eicholz wrote:
> I don't know of any cases that you are referring to here

Arron, I was referring to these testcases


which I mentioned and commented in my January 25th 2010 email with
subject line:
t1508-c527-font-00-b.htm to t1508-c527-font-10-c.htm: Testcases pass
condition are not self-evident


> but I would be
> happy to address those cases if you call them out (note I haven't gotten
> a chance to address any issues since 2/9 so I am a bit behind on
> feedback if you already called them out).

No problem. I will call them out when I encounter them.

>> > Plus, we should have images that are exact measurements.  One for
>> mm,
>> > one for inches, etc.  So if an end result of a test is supposed to
>> > have a space of 10mm, the tester can clearly see that the end
>> results
>> > are 10 mm.
>> Creating custom (vertical and horizontal) measurement rulers (for in,
>> mm,
>> cm, px, etc.) is rather easy to do. I have submitted 2 for the test
>> suite.
> Rulers are a great idea but you forget a couple major issues with using
> images. Zooming and HighDPI. These two features do not have any
> guidelines on how they should react on images. Are they required to
> scale the images? I say yes but that then can possibly change image
> rendering a 1px line into 2px or more. That is problematic. Using rulers
> restricts us from testing zooming and HighDPI scenarios with the same
> tests that we have already created. This is one reason why the '96dpi'
> flag was created because of this particular issue. That flag defines
> that the tests is only valid in that specific DPI setting and may not
> work in other DPI settings.

I try to follow as strictly as possible the assumptions for the CSS 2.1
test suite. So,

Zooming: I never use this, never test this, never bother with such
feature and there is nothing mentioned about zooming in the test suite
assumptions. I am not sure CSS 2.1 testsuite testcases should still be
workable and reliable if zoom feature (magnification of the whole
webpage, not just text size increase) is activated.

HighDPI: there is now 8777 test entries in the test suite now. I
personally only test with 96dpi. If we have to start testing with
zooming and/or with various dpi settings, then such kind of testing will
require appropriate software for such kind of testing.

>> What is unacceptable is that a few testcases expect user agents to be
>> able to
>> render distance like 2.54mm (9.6px) or 15.24mm (or 57.6px) precisely
>> and
>> then ask/invite the tester to check if there is red in the page.
>> There is no normative restraint or requirement regarding how fraction
>> of a
>> pixel is supposed to be handled by user agents. Each/all user agents
>> have
>> implementation limits and various capability limitations.
> If there are cases that have this fractional pixel problem let me know.

If I see another testcase with such fractional pixel problem, I will let
you know.

> I know I have been trying to track all of them down and fix them.

Yes, Arron, you have.

is one example which was reported in
and it was corrected in

Regarding the 57.6px example, please consult


and look for "top: 0.6in;" in



One last comment. Specifically regarding Microsoft's line-height
testcases: there are lots of testcases inviting the tester to compare 2
small black boxes' height and then establish if the 2 small black boxes
are the same height (e.g.:
I personally believe that those testcases are definitely not ideal.

best regards, Gérard Talbot
Contributions to the CSS 2.1 test suite:

CSS 2.1 test suite (alpha 1; January 27th 2010):

CSS 2.1 test suite contributors:
Received on Tuesday, 2 March 2010 18:51:09 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 21:13:19 UTC