On Mon, 14 Dec 2009, Andreas Raab wrote:

> Levente Uzonyi wrote:
>> I'd like to have less failing tests in the trunk. For this, we have to make 
>> decisions/anwser questions.
>
> Thanks for taking the time to go through these tests carefully!
>
>> Should we change the default, so that LongTestCases are run when selected 
>> in the Test Runner? (Or one could add a chechbox to the gui which would let 
>> the user to decide if she/he wants to run long tests? Or this could be a 
>> preference, etc.)
>
> I would opt for a preference or a checkbox in the UI.

I added a preference, maybe someone will update the gui.

>
>> Mirror primitives do not exists in the current vm. Will the vm have mirror 
>> primitives? Should we keep the tests? Removing these would "fix" 6 errors.
>
> IIRC, the mirror primitive tests were designed to test that certain 
> primitives will be applied to the "last argument" instead of only to the 
> receiver. I.e., such that one could do both:
>
> Object>>class
>  "answer the class of the receiver"
>  <primitive: 1234>
>
> as well as:
>
> Mirror>>classOf: anObject
>  <primitive: 1234>
>
> It would be good if we could keep those working (perhaps ask Eliot for an 
> opinion).
>

We should definitely keep the tests. After a bit of googling I found 
Eliot's email which didn't get much attention: 
http://lists.squeakfoundation.org/pipermail/vm-dev/2009-September/003161.html
This explains everything about mirror primitives. I wonder why these 
changes aren't integrated into VMMaker. (Note that ContextPart >> 
#objectClass: is missing from the attached source, I added an 
implementation to the trunk). I marked the failing tests as expected 
failures.

>> Building a vm with the latest VMMaker lets 6 failing test pass 
>> (AllocationTest, BitBltTests). It also avoids crashing linux as reported 
>> before
>> (http://lists.squeakfoundation.org/pipermail/squeak-dev/2009-December/142074.html). 
>
> I would opt to fail the crashing AllocationTests on Linux, i.e., something 
> like:
>
>    "the test currently crashes the vm on unix"
>    Smalltalk platformName = 'unix' ifTrue:[^self assert: false].
>
> mark it expected failure, and later, when we have fixed VMs, allow the test 
> to run.
>

Done, feedback is welcome.

>> And it also yields slight performance improvement because of the 
>> primitiveNextPut fix. Why aren't vm releases available with these fixes?
>
> Because nobody's been asking for then :-) A post to vm-dev might help.
>

I'm on it.

>> There are two "great challenges" which would let most other failing tests 
>> pass:
>> - fixing traits (the issues are solved in pharo, but one has to extract and 
>> integrate them)
>> - fixing the decompiler (there's an issue with temporary variable 
>> declarations in inlined blocks)
>> Is anybody working on these?
>
> I'm working on the former (time permitting) and I'd check with Eliot for the 
> latter.
>

Great. :)

>> - MorphicToolBuilderTests>>#testGetButtonSideEffectFree
>
> I can take care of that.
>

Cool. :)


Levente

> Cheers,
>  - Andreas
>
>