[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Public WebGL] Bug in conformance/glsl/misc/shader-with-line-directive.html

Thanks for the clarification. Actually, upon checking, I found that
this test was already excluded from the run lists in both the 1.0.1
and top of tree conformance suites, precisely because it made
assertions about error messages not guaranteed by the GLSL spec. I've
deleted the invalid test and test list entries.


On Tue, Jul 10, 2012 at 11:23 AM, Gregg Tavares (社用) <gman@google.com> wrote:
> (d) do something else.
> remove the test.
> It's a bogus test (sorry). GL is not required to provide any error messages
> what-so-ever. All it's required to do is fail invalid programs. So the line
> directive is pointless and a test for it is pointless as well. My bad, I
> didn't know that at the time I wrote the test.
> On Tue, Jul 10, 2012 at 11:14 AM, Kenneth Russell <kbr@google.com> wrote:
>> Alok Priyadarshi from the Chrome team has been working for several months
>> on a brand new preprocessor for the ANGLE shader translator. This new
>> preprocessor has been written from the ground up and has hundreds of
>> associated unit tests. It has recently been turned on in Chromium builds by
>> default. The old preprocessor in ANGLE has historically been one of the
>> buggiest areas of the compiler, so Alok's rewrite is both timely and
>> welcome.
>> The new preprocessor turned up a bug in the
>> conformance/glsl/misc/shader-with-line-directive.html test. The test
>> incorrectly assumes that the error will occur at line 124, when it should
>> happen at line 123. See
>> http://code.google.com/p/chromium/issues/detail?id=136441 .
>> I will update the top of tree test to be correct. This will cause it to
>> start failing in Firefox and Safari until ANGLE is updated there, and
>> presumably in Opera as well (unless Opera's preprocessor has always been
>> correct, and it's been failing this test all along).
>> The question is what to do about the 1.0.1 version of the test. The test
>> is wrong, but many implementations have been incorrectly passing the buggy
>> version of the test. Should we:
>>   (a) update the test to be correct?
>>   (b) disable the test in the 1.0.1 conformance suite, so as not to
>> retroactively add a new requirement?
>>   (c) do both (a) and (b)?
>>   (d) do something else?
>> Thanks for your feedback,
>> -Ken

You are currently subscribed to public_webgl@khronos.org.
To unsubscribe, send an email to majordomo@khronos.org with
the following command in the body of your email:
unsubscribe public_webgl