Re: Recent GCPRO removal and copying GC

[Paul Eggert]

On 09/18/2015 12:31 PM, Pip Cet wrote:
> If I understand things correctly, such a garbage collector must know 
> precisely which stack locations correspond to Lisp objects and which 
> ones don't

Yes and no.  It's reasonably common to use Bartlett's idea of a 
mostly-copying collector, in which stack locations and registers are 
ambiguous and therefore pin down objects (which can't be moved) but 
managed references don't have that problem and in practice most objects 
can be moved.  I think the Chakra and WebKit JavaScript engines are both 
like that.  For more, please see:

Shahriyar R, Blakburn SM, McKinley KS. Fast conservative garbage 
collection. Proc OOPSLA '14, 121-139. 
http://dx.doi.org/10.1145/2714064.2660198

Getting back to your main point, the GCPROs weren't being maintained 
well any more as they weren't being tested.  If any new garbage 
collector needs them we can bring them back.  I hope this doesn't 
happen, as they were a real obstacle to maintenance, not only for 
existing developers but also for potential new recruits.
<http://splashcon.org>