Hi Chris,

> On Mar 25, 2018, at 9:16 PM, Chris Cunningham <cunningham.cb at gmail.com> wrote:
> 
> 
> 
> On Mar 25, 2018 3:56 PM, "Eliot Miranda" <eliot.miranda at gmail.com> wrote:
> Hi Chris,
> 
> On Sun, Mar 25, 2018 at 1:51 PM, <commits at source.squeak.org> wrote:
>> Chris Cunningham uploaded a new version of KernelTests to project The Trunk:
>> http://source.squeak.org/trunk/KernelTests-cbc.336.mcz
>> 
>> ==================== Summary ====================
>> 
>> Name: KernelTests-cbc.336
>> Author: cbc
>> Time: 25 March 2018, 1:51:33.345291 pm
>> UUID: 6e6b91ac-d706-8f4d-bd60-1057109994cb
>> Ancestors: KernelTests-eem.335
>> 
>> Test that new xor: takes blocks - but only those that result in booleans.  Others raise errors.
>> 
>> =============== Diff against KernelTests-eem.335 ===============
>> 
>> Item was changed:
>>   ----- Method: FalseTest>>testXor (in category 'tests') -----
>>   testXor
>>         self assert: (false xor: true) = true.
>>         self assert: (false xor: false) = false.
>> +       self assert: (false xor: [true]) = true.
>> +       self assert: (false xor: [false]) = false.
>> +       self should: [false xor: [1]] raise: NonBooleanReceiver.!
> 
> Shouldn't "false xor: 1" also raise an error? 
> Yes it will. I will put a check for that in I'd it makes sense. 
> I don't see why "false xor: [1]" should raise NonBooleanReceiver specifically.
> That isn't ideal. It is a result of the implementation - ifTrue:ifFalse: is sent to the (value of the) argument which is then the "receiver" at that point and isn't an boolean.
>   It should raise an error, but whether [1] or 1 becomes the receiver internally to xor: seems to me an implementation detail.  Surely the receiver in "false xor: [1]" is false and that ids a boolean, so NonBooleanReceiver is an unintuitive error foBoolean.
> Yes but trapping this and making it more intuitive would make the xor: logic even more complex. 

You misunderstand.  The test case should test for Error, not for NonBooleanReceiver.  Think of the tests as specifications, not precise descriptions of current behaviour.  Yes, the current /implementation/ raises NonBooleanReceiver (which is an error and hence meets the specification), but the specification requires only that aBoolean xor:  anything other than a Boolean or a valuable (something that understands value) that evaluates to a Boolean is an error.

> 
>  
>> -
>> -       self
>> -               should: [(false xor: [false])
>> -                       ifTrue: ["This should never be true, do not signal an Error and let the test fail"]
>> -                       ifFalse: [self error: 'OK, this should be false, raise an Error']]
>> -               raise: Error
>> -               description: 'a Block argument is not allowed. If it were, answer would be false'.!
>> 
>> Item was changed:
>>   ----- Method: TrueTest>>testXor (in category 'testing') -----
>>   testXor
>>         self assert: (true xor: true) = false.
>>         self assert: (true xor: false) = true.
>> +       self assert: (true xor: [true]) = false.
>> +       self assert: (true xor: [false]) = true.
>> +       self should: [true xor: [1]] raise: NonBooleanReceiver.!
>> -
>> -       self
>> -               should: [(true xor: [true])
>> -                       ifTrue: ["This should never be true, do not signal an Error and let the test fail"]
>> -                       ifFalse: [self error: 'OK, this should be false, raise an Error']]
>> -               raise: Error
>> -               description: 'a Block argument is not allowed. If it were, answer would be false'.!
>> 
>> 
> 
> 
> 
> -- 
> _,,,^..^,,,_
> best, Eliot
> 
> 
> 
> 
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20180326/63b488b4/attachment.html>