[squeak-dev] The Trunk: KernelTests-dtl.235.mcz

Frank Shearar frank.shearar at gmail.com
Mon Nov 5 13:21:59 UTC 2012


On 5 November 2012 13:17, David T. Lewis <lewis at mail.msen.com> wrote:
> On Mon, Nov 05, 2012 at 12:21:27PM +0000, Frank Shearar wrote:
>> On 5 November 2012 11:44, David T. Lewis <lewis at mail.msen.com> wrote:
>>
>> > 3) LargeNegativeIntegerTest.testMinimumNegativeIntegerArithmetic
>> >
>> > This is a VM capability issue (passes on VMs with the necessary update).
>> > It is not something that can or should be addressed in the image, therefore
>> > IMO it should not block the 4.4 release.
>>
>> Ideally we'd have it an expected failure on a Cog VM and passing in
>> the newer interpreter VMs, but I agree that we shouldn't delay 4.4 any
>> further.
>
> This is not a Cog issue, it is an issue that affects all VMs. It is a
> real failure (not expected) and it will go away when all VMs have been
> updated.  Meantime, let's just ensure that the test is green when run on
> a VM (Cog or interpreter) that *does* have the necessary fix.

I say "on a Cog VM" because the Cog VM in the squeak-ci repository
(which is, IIRC, the latest released version) drives the CI tests and
doesn't have the fix. Thus, if a random newbie downloads Cog (of the
same version that CI uses) and the latest image, they will see a red
bar when they run this test. They won't know or care about the details
of why this test passes on this VM or fails on that. That's a problem.

But you do point out below that the test at least says why it would/might fail.

frank

> For background, here is the explanation in the test comment:
>
> testMinimumNegativeIntegerArithmetic
>         "We are speaking of minimum integer in underlying hardware here.
>         In 2-complement, abs(INT_MIN) = (INT-MAX+1) and thus overflows hardware register.
>         Since some old VM forgot this edge case they may fail and it's better to be aware of it.
>         http://code.google.com/p/cog/issues/detail?id=92
>         http://bugs.squeak.org/view.php?id=7705
>         We only test the cases of 32 and 64 bit signed integers."
>
> BTW, kudos to Nicolas for finding and fixing this :)
>
> Dave
>
>


More information about the Squeak-dev mailing list