On Mon, 23 Aug 2010, Nicolas Cellier wrote:

> 2010/8/23 Levente Uzonyi <leves at elte.hu>:
>> On Mon, 23 Aug 2010, Nicolas Cellier wrote:
>>
>>> Waouh!
>>>
>>> I don't know how you could possibly find this one!
>>> Chapeau bas
>>
>> It was just an "accident". There was a discussion on the Pharo list about
>> converting a ByteArray to an Integer. Since LargeIntegers are
>> variableByteSubclasses (just like ByteArrays), I tried this:
>> #[1] as: LargePositiveInteger.
>> The printString was empty without #normalize, so I started debugging. It's a
>> bit strange though, that LargePositiveInteger >> #printOn:base: has the
>> following lines:
>> "Don't engage any arithmetic if not normalized"
>> (self digitLength = 0 or: [(self digitAt: self digitLength) = 0]) ifTrue:
>> [^self normalize printOn: aStream base: b].
>> But the "artifically created" not normalized LargePositiveInteger passes
>> this check.
>>
>
> Ah, I see, nice trick...
> Does it deserve a dedicated test, or is it too fragile a feature ?

I think that we shouldn't expect that not LargeIntegers work correctly 
without normalization. For example:
(#[1] as: LargePositiveInteger) = 1 "===> false"
So I think this shouldn't be tested.


Levente

>
> Nicolas
>
>>
>> Levente
>>
>>>
>>> Nicolas
>>>
>>> 2010/8/23  <commits at source.squeak.org>:
>>>>
>>>> Levente Uzonyi uploaded a new version of Kernel to project The Trunk:
>>>> http://source.squeak.org/trunk/Kernel-ul.485.mcz
>>>>
>>>> ==================== Summary ====================
>>>>
>>>> Name: Kernel-ul.485
>>>> Author: ul
>>>> Time: 23 August 2010, 2:13:52.357 pm
>>>> UUID: fce08992-f821-ed46-a863-9935d6a7fe2e
>>>> Ancestors: Kernel-dtl.484
>>>>
>>>> - fix LargeInteger printing
>>>>
>>>> =============== Diff against Kernel-dtl.484 ===============
>>>>
>>>> Item was changed:
>>>>  ----- Method: LargePositiveInteger>>printOn:base:nDigits: (in category
>>>> 'printing') -----
>>>>  printOn: aStream base: b nDigits: n
>>>>        "Append a representation of this number in base b on aStream using
>>>> n digits.
>>>>        In order to reduce cost of LargePositiveInteger ops, split the
>>>> number of digts approximatily in two
>>>>        Should be invoked with: 0 <= self < (b raisedToInteger: n)"
>>>>
>>>>        | halfPower half head tail |
>>>>        n <= 1 ifTrue: [
>>>>                n <= 0 ifTrue: [self error: 'Number of digits n should be
>>>>> 0'].
>>>>
>>>>                "Note: this is to stop an infinite loop if one ever
>>>> attempts to print with a huge base
>>>>                This can happen because choice was to not hardcode any
>>>> limit for base b
>>>>                We let Character>>#digitValue: fail"
>>>> +               ^aStream nextPut: (Character digitValue: self) ].
>>>> -               ^Character digitValue: self].
>>>>        halfPower := n bitShift: -1.
>>>>        half := b raisedToInteger: halfPower.
>>>>        head := self quo: half.
>>>>        tail := self - (head * half).
>>>>        head printOn: aStream base: b nDigits: n - halfPower.
>>>>        tail printOn: aStream base: b nDigits: halfPower!
>>>>
>>>>
>>>>
>>>
>>
>>
>>
>>
>
>