Re: [Sbcl-devel] more callback fun

[<me...@ho...>]

On Wednesday 04 January 2006 23:22, Cyrus Harmon wrote:
> Well, while I wait for sf.net's SBCL anoncvs to work again, I can ask
> a question about callbacks. There's a FIXME about the SAPs not being
> GC-safe. This sounds plausible to me, but I'm not sure what the best
> fix is. Anyone have a suggestion for making sure args-sap and res-sap
> don't move? Wrap the whole thing in a without-gcing block? Do we care
> about making this GC-safe? It would seem so, but I could be wrong...

I put that note there some time ago and still think it's true. But, it=20
could be hard to trigger, because it needs a) another thread b) an=20
interrupt to initiate gc after get-lisp-obj-address but before the sap=20
is made.

On conservative gc platforms it may not be possible to trigger it as the=20
lisp-obj-address is in a register or on the stack. With=20
non-conservative gc an interrupt at that point could be fatal.

The hypotethical non-conservative threading platform would be especially=20
vulnerable.

WITHOUT-GCING is quite expensive due to the unwind-protect in it. This=20
may be a concern. I think WITH-PINNED-OBJECTS is right fix. But, sadly=20
on cheneygc it is equivalent to without-gcing.

G=E1bor

>
> Thanks,
>
> Cyrus
>
>
>
>
> -------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc. Do you grep through
> log files for problems?  Stop!  Download the new AJAX search engine
> that makes searching your log files as easy as surfing the  web.=20
> DOWNLOAD SPLUNK!
> http://ads.osdn.com/?ad_id=3D7637&alloc_id=3D16865&op=3Dclick
> _______________________________________________
> Sbcl-devel mailing list
> Sbc...@li...
> https://lists.sourceforge.net/lists/listinfo/sbcl-devel