Re: redoing SCM representation in 2.2

[Andy Wingo]

On Sun 15 May 2011 11:02, Ken Raeburn <address@hidden> writes:

> Is that really any more of an issue this way than with the current
> encoding -- if not for SCM, then for heap data structures including both
> SCM objects and integers or characters?  I thought the GC code already
> had to cope with things looking like they could be pointers but not
> actually corresponding to anything allocated via the GC library.

It's not an issue for SCM values.  The Scheme stack has only Scheme
values, so there's no problem there either; and the C stack is small,
mostly ephemeral, and there is good blacklisting, so no problem there
either, AFAIK.  Heap values which contain no GC-controlled pointers are
allocated "atomically", which means the GC doesn't trace them.  So the
issue would only be with structures which are traced by GC and include
both, and there are very few of those.

Regards,

Andy
-- 
http://wingolog.org/