Re: GC: cons sweeping and cons block size

[Ken Raeburn]

On Jul 5, 2007, at 12:02, Stefan Monnier wrote:
(quoting David, who was quoting Stefan)
> > > I haven't found the time to do it, but it would be good to get  
> > > rid of this
> > > situation and always place tags in the 3 LSB bits.

It does appear to be the default for most systems, maybe all modern  
ones, but only when we compile with GCC.  Are we confident no one is  
using Emacs on systems where we use a system malloc that doesn't do 8- 
byte alignment?  Or where the native compiler doesn't do 8-byte  
alignment for static objects?  I think I'd be happier with changing  
the default to use USE_LSB_TAG always but check the malloc return  
values and compiler-produced alignment on the configurations where we  
don't already know we're safe, and perhaps delete that code and maybe  
much of the non-USE_LSB_TAG code a while after it's released without  
related bug reports; I'm afraid that temporarily increases the  
ugliness rather than reducing it though.

> Luckily, YAILOM problems can be cought by the compiler if you
> use -DUSE_LISP_UNION_TYPE (but it incurs a performance hit, which  
> is why
> it's not used by default).

Hm, I don't think I ever got around to seeing if __attribute__ 
((transparent_union)) would fix the performance...

We could do a USE_LSB_TAG variant of the union, to get more address  
space back but keep the paranoid type checking capability.

Ken