Re: new implementation of bitmap boolean operations

[Paolo Bonzini <bonzini at gnu dot org>]

> > rather than optimizing this more and more, I'd look into using 
> > Michael Hayes' bitset library that is already used by bison; this is 
> > made much easier by your previous patches to use different names for 
> > the various binary operations.
>
> thanks for that pointer.  I don't think inserting that library is a stage
> 3 task.

Yes, of course.  And I also do not know how much this has to do with 
CodeSourcery contracts (I suspect a lot...); but I don't know how much 
room there is to gain now; optimized iterators and 
equal_p/empty_p/intersect_p were surely the biggest thing.  Bitmaps are 
easy to overoptimize :-)

> It does provide a nice implementation abstraction -- replacing
> compile time type info with rtti.  Is that a win overall? I don't know.

Neither do I.  But I think that sometimes we are hitting the barriers of 
the current bitmap.c and papering elsewhere (see 
http://gcc.gnu.org/ml/gcc-patches/2004-09/msg02583.html for example).

Paolo