[squeak-dev] Re: [ANN] Kernel methods license audit

Paolo Bonzini bonzini at gnu.org
Fri Mar 21 08:26:08 UTC 2008


> However, looking at the source, apart accessors, some methods will 
> hardly ever change, like:
> Integer>>bitInvert
>     ^-1-self
> 
> There is no added value in such method but very obvious algorithm:
> two complement: bitinvert the digits, and add 1
> => bitInvert: take two complement, and subtract 1

This could become "^self bitXor: -1", at least if bitXor does not use 
bitInvert. :-)

> The trick is not here, it is in pretending LargeNegativeInteger are 
> stored in two complements, what they are not...
> So the trick is in all the bitOr:, bitAnd:, bitXor:, bitShift: 
> implementations, as a whole.

If you want a >0% chance to have different code, look at libgst/mpz.c 
from GNU Smalltalk.  It contains the exact same algorithms written in C 
(so don't be afraid of tainting anything with anything).  And XOR does 
not use NOT so you get a chance to rewrite #bitInvert too.  :-)

Paolo



More information about the Squeak-dev mailing list