I know some of them:

(Integer new: 5 neg: true) class.
(SmallInteger maxVal + 1 replaceFrom: 4 to: 4 with: #[0] startingAt: 1)
class.
(SmallInteger maxVal + 1 digitDiv: 4 neg: false) first class.
(SmallInteger maxVal + 1 \\\ 3) class.

And probably a few others (digitRshift:bytes:lookfirst:).
IMO, it's the responsibility of senders to normalize, and we should clearly
indicate the comment of each of these messages.
I also think that the last one, \\\, should send normalize, because it won't
change performance of senders noticeably.

Nicolas

2011/7/11 Eliot Miranda <eliot.miranda at gmail.com>

> Hi Igor,
>
> On Mon, Jul 11, 2011 at 4:30 AM, Igor Stasenko <siguctua at gmail.com> wrote:
>
>> Levente, your change is a bit dangerous,
>> because if i remember correctly, not all primitives normalizing large
>> integers.
>> So, it is fairly possible to have large integer with value in a range
>> of small integer.
>>
>
> It is difficult to construct such a beast.  If one uses the
> interpreterProxy API, positive32BitIntegerFor et al you'll always obtain
> something normalized.  AFAIA, the only way one can create something
> unnormalized is by asking for a large integer and then altering its
> contents. So for me this isn't a dangerous change and requiring VM code to
> always deal with normalized integers seems rational and dependable to me.
>
>
>> Not saying , that one can manipulate largeinteger bits directly, and
>> then use it later in comparison,
>> which will fail due to your changes.
>> So, such change requiring strong guarantees that you cannot create
>> large integers which is in range of small integers.
>>
>> On 10 July 2011 16:36,  <commits at source.squeak.org> wrote:
>> > Levente Uzonyi uploaded a new version of Kernel to project The Trunk:
>> > http://source.squeak.org/trunk/Kernel-ul.604.mcz
>> >
>> > ==================== Summary ====================
>> >
>> > Name: Kernel-ul.604
>> > Author: ul
>> > Time: 10 July 2011, 4:36:20.467 pm
>> > UUID: f758876f-a507-404b-aa2f-2c1038504b48
>> > Ancestors: Kernel-bf.603
>> >
>> > Optimized Integer >> #= for integer-integer comparisons. Return false
>> for integers from different classes without further checking.
>> >
>> > =============== Diff against Kernel-bf.603 ===============
>> >
>> > Item was changed:
>> >  ----- Method: Integer>>= (in category 'comparing') -----
>> >  = aNumber
>> > +
>> > +       aNumber isInteger ifTrue: [
>> > +               aNumber class == self class ifFalse: [ ^false ].
>> > +               ^(self digitCompare: aNumber) = 0 ].
>> > +       aNumber isNumber ifFalse: [ ^false ].
>> > +       ^aNumber adaptToInteger: self andCompare: #=!
>> > -       aNumber isNumber ifFalse: [^ false].
>> > -       aNumber isInteger ifTrue:
>> > -               [aNumber negative == self negative
>> > -                       ifTrue: [^ (self digitCompare: aNumber) = 0]
>> > -                       ifFalse: [^ false]].
>> > -       ^ aNumber adaptToInteger: self andCompare: #=!
>> >
>> >
>> >
>>
>>
>>
>> --
>> Best regards,
>> Igor Stasenko AKA sig.
>>
>>
>
>
> --
> best,
> Eliot
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20110711/b2b7aeec/attachment.htm