[Gcl-devel] possible type_of acceleration

[Camm Maguire]

Greetings!

As we all know, type_of has gotten more expensive in support of
immediate fixnums and two word conses.  I have a likely silly idea I'd
like to float in this regard.

It appears we really want to detect cons fast.  What if pages were
only allocated in pairs, the first of which is any type, and the
second of which is always cons?  That way we can read a cons off by
its address with no memory indirection.  Disadvantage, of course is
that exactly half the heap is always cons.  We could relax this a bit
in a hybrid strategy.  Is this asking for trouble?

BTW, Paul, is there a complete set of functions listed anywhere that
need be able to handle an improper list, so that I might dispense with
the old DOTNIL trick?


Take care,

Robert Boyer <address@hidden> writes:

> > your ideas on when a t6 and later tag might be useful
> > (i.e. when things are acceptable) are most welcome.
> 
> I'm quite happy to just grab 2.7.0 and take my chances.  I keep the last
> version of 2.7.0 I have grabbed that worked around just in case the new one
> is broken.
> 
> Thanks,
> 
> Bob
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah