VM Bug

I have been using Pharo-2.0's one-click cog VM with Pharo.

There appears to be a bug in ShortIntegerArray on Linux (not tested on other platforms) such that you can't store 16-bit values iff the high bit is set.

Not a problem for most of you, but I was using ShortIntegerArray-s to hold Unicode code points.

http://code.google.com/p/pharo/issues/detail?id=7569

FYI,
-KenD
--
Ken [dot] Dickey [at] whidbey [dot] com

_______________________________________________
Cuis mailing list
[hidden email]
http://jvuletich.org/mailman/listinfo/cuis_jvuletich.org

Hi Ken,

On 2/22/2013 2:04 AM, Ken Dickey wrote:
Cuis with Cog on my Mac shows same behavior. So I digged a little bit.
It is not a bug. ShortIntegerArray handles 16-bit unsigned integers. I
corrected the class comment. I.e.:

"ShortIntegerArray : unsigned"
sia := ShortIntegerArray new: 8 withAll: 0.
sia at: 1 put: 16r8421.        "fails"
sia at: 1 put: -16r8000.    "ok"
sia at: 1.

"IntegerArray : signed"
ia := IntegerArray new: 8 withAll: 0.
ia at: 1 put: 16r84210000.      "Converts to negative"
ia at: 1.
ia at: 1 put: -16r80000000.    "ok"
ia at: 1.

"WordArray: unsigned"
wa := WordArray new: 8 withAll: 0.
wa at: 1 put: 16r8421.                    "ok"
wa := WordArray new: 8 withAll: 0.
wa at: 1 put: -1.

Note that WordArray stores unsigned numbers, but the primitives handle
signed. If you want unsigned 16 bit, I'd rather add #unsignedAt: and
#unsignedAt:put: to ShortIntegerArray, doing a conversion like WordArray
does.

Cheers,
Juan Vuletich

_______________________________________________
Cuis mailing list
[hidden email]
http://jvuletich.org/mailman/listinfo/cuis_jvuletich.org

On Fri, 22 Feb 2013 09:20:54 -0300
Juan Vuletich <[hidden email]> wrote:

> Note that WordArray stores unsigned numbers, but the primitives handle
> signed. If you want unsigned 16 bit, I'd rather add #unsignedAt: and
> #unsignedAt:put: to ShortIntegerArray, doing a conversion like WordArray
> does.

Yes.  I want to store all 16 bits.  Do whatever is the sensible thing.

The comment in #at:put: is definitely at odds with what the primitives do.

My personal preference is for unsigned byte, word, and long storage to have better naming (e.g. U8Array, U16Array, U32Array with #at:put: storing unsigned integers to  16rFF*) and have these subclassed for (S8Array, S16Array, S32Array) with #at:put: expecting signed integers of the proper size.

Number implementations are way too complex and should be explained better for beginners.

Thanks for the digging,
-KenD
--
Ken [dot] Dickey [at] whidbey [dot] com

_______________________________________________
Cuis mailing list
[hidden email]
http://jvuletich.org/mailman/listinfo/cuis_jvuletich.org

Hi Ken,

On 2/22/2013 12:56 PM, Ken Dickey wrote:
> On Fri, 22 Feb 2013 09:20:54 -0300
> Juan Vuletich<[hidden email]>  wrote:
>
>> Note that WordArray stores unsigned numbers, but the primitives handle
>> signed. If you want unsigned 16 bit, I'd rather add #unsignedAt: and
>> #unsignedAt:put: to ShortIntegerArray, doing a conversion like WordArray
>> does.
> Yes.  I want to store all 16 bits.  Do whatever is the sensible thing.

I added new protocol to ShortIntegerArray and enhanced comments in this
and related classes. Update is at GitHub.

> The comment in #at:put: is definitely at odds with what the primitives do.
>
> My personal preference is for unsigned byte, word, and long storage to have better naming (e.g. U8Array, U16Array, U32Array with #at:put: storing unsigned integers to  16rFF*) and have these subclassed for (S8Array, S16Array, S32Array) with #at:put: expecting signed integers of the proper size.

Sounds reasonable. Didn't implement this, though.

> Number implementations are way too complex and should be explained better for beginners.
>
> Thanks for the digging,
> -KenD

Cheers,
Juan Vuletich

_______________________________________________
Cuis mailing list
[hidden email]
http://jvuletich.org/mailman/listinfo/cuis_jvuletich.org