optimization

[Manuel Guesdon]

Hi,

Would it be possible to add public declaration of boolN and boolY in Additions, 
with something like this:
#ifdef NO_GNUSTEP
#define NSNumberYes [NSNumber numberWithBool:YES]
#define NSNumberNo [NSNumber numberWithBool:NO]
#else
#define NSNumberYes boolY       //Or another name
#define NSNumberNo boolN       //Or another name
#endif

So we won't have to call -numberWIthBool: each time we want a bool number.

Is it also possible to declare function wich use IMP for NSNumber 
numberWithInt:, 
numberWithFloat:,..
like this:

#ifdef NO_GNUSTEP
#define NSNumberWithInt(value) [NSNumber numberWithInt:value]
...
#else
#define NSNumberWithInt(value) 
((*methodIMP)(NSNumberClass,NSNumberWithIntSelector,value))
...
#endif

This probably save a lot of objc_msgSend calls with no (?) drawbacks (except 
poseAs: and dirty 
class methods manipulations).



And my favorite one, a little more complex:
"How to store millions NSIntNumber in 0 byte space" :-)

Using objects to store int numbers can take a lot of memory (real value 
storage+class 
pointer+retain count,...) and cost allocation/deallocation/retain/release time

If we say that upper addressable memory won't be used (physicall memory will 
be less than addressable memory).
Say we 'reserve' (not allocate) a zone of 1024x1024=intNb starting at (unsigned 
char*)intBase 
and stopping at (unsigned char*)intLimit in this zone

we could have:
NSNumber -numberWithInt:(int)value like this
{
        if (value>=0 and value<intLimit)
                return ((unsigned char*)value)+intBase;
        else
        {
                usual allocation
        }
}

and -(int)intValue like that:
{
        if (self>=0 and self<intLimit)
                return ((unsigned char*)self)-intBase;
        else
        {
                usual allocation
        }
}
we only have to modify NSNumber class to handle this case and some 
other functions/macros (obj_msgSend,object_get_class,...).

cost: 2 tests in some functions
gain: no allocation/deallocation/retain/release and no physical memory 
usage for the choosen int range
drawback(s): some tools can't check this memory and may report 
errors.

What do you think ? Am I crazy ? :-)


Manuel



--
______________________________________________________________________
Manuel Guesdon - OXYMIUM <address@hidden>
14 rue Jean-Baptiste Clement  -  93200 Saint-Denis  -  France
Tel: +33 1 4940 0999  -  Fax: +33 1 4940 0998