Re: Collecting markers with MPS

[Gerd Möllmann]

Helmut Eller <eller.helmut@gmail.com> writes:

> On Wed, Apr 24 2024, Gerd Möllmann wrote:
>
>> Yes, it makes sense. One could of course use a timer. Or maybe, don't
>> know, run something when we handle an MPS message. That just occurred to
>> me a second ago, so it might be nonsense.
>
> A variation could be this: we split the marker type up into three parts.
> One part is only handled by the C code (malloc-ed, non-moving,
> reference-counted), the second part is handed out to Lisp, and the third
> part is specifically for the undo-list.
>
> The Lisp part is basically a pointer to the C part, but it has a
> finalizer that decrements the reference count in the C part.
>
> The third part is also a pointer to the C part, but the C part also
> keeps a (traced) reference to this part.  When the reference count drops
> to zero, we can set the pointer from the undo-list to the C part so 0.
> Ie. this is basically weak reference, implemented manually.
>
> An advantage of this scheme is that the GC only needs to support
> finalization.  Also, the involved changes would likely work for both the
> old GC and the new GC.  Which might be an issue if the new GC remains a
> purely optional feature.

Hm, I must say that I like your weak-refs a _lot_ better. I mean a LOT.
(Google translation of weak ref = schwacher Schiedsrichter, BTW :-)).
And I think making a weak-ref for the old GC is also be relatively easy.