Re: [Help-smalltalk] gst-shape

[Paolo Bonzini]

On 10/18/2010 11:42 AM, Gwenaël Casaccio wrote:
> Hi,
>
> I would like to announce gst-shape git://github.com/MrGwen/gst-shape.git
> a refactoring of the class shape, a small example of the power of this new
> refactoring:

This is very nice, thanks.  You can also use a branch of smalltalk.git 
rather than a fully separate repository.  This will make it easier for 
me to pull.

However, I wouldn't use class methods.  I think it's bad design. It's 
better to use a simpler hierarchy and put these on the instance side 
using instance variables:

Shape "class methods for mapping integers to ShapeSymbol instance"
    VMShape
        FixedShape "current NilShape"
        IntegerShape "adds #byteSize, #byteSize:"
            SignedIntegerShape
            UnsignedIntegerShape
        FloatingPointShape
        CharacterShape
             EncodedCharacterShape
             UnicodeCharacterShape
    ShapeAliasSymbol "abstract class maybe not necessary"
        ShapeInherit
        ShapeWord

This is because you can use #nextBytes:signed: and #nextPutBytes:of: to 
implement dumping.  Those methods are private, but we can un-privatize them.

"wrongClassIfFloatOrDoubleShape:" and 
"wrongClassIfCharacterOrUtf32Shape:" can be unified to 
"checkArgumentClass:".

The "subclassResponsibilityIfCharacterOrUtf32Shape" and 
"subclassResponsibilityIfNotCharacterOrUtf32Shape" methods are also bad. 
 First of all, the former should not be used (it's always using 
#ifFalse: in kernel/CharArray.st). Second, I'd anyway use a method 
#hasValueAtPrimitives that returns true/false.

Thanks,