Re: Opportunistic GC

[Pip Cet]

On Mon, Mar 8, 2021 at 11:38 AM Stefan Kangas <stefankangas@gmail.com> wrote:
> Pip Cet <pipcet@gmail.com> writes:
> > Note that none of this is "real" GC: we still mark and sweep, just in
> > a slightly smarter way.
>
> Do you see any significant pros (or cons) to the approach you suggest as
> compared to a generational GC?

In general, I see significant cons in my approach of incrementally
improving GC a little here and there: a complete rewrite would be much
better (but, at least at this point in time, I'm not volunteering for
it). So I think any time sunk into this may be lost, eventually, and
we're only patching holes in our straw hut until we find time to
demolish it and build a skyscraper instead.

Other cons:
- requires POSIX
- requires efficient OS paging

Pros:
- it's cheap, typically (i.e. on GNU/Linux with a "typical" workload)
- it's easy to do (comparatively)
- there are going to be tangible and immediate benefits
- it's a first step towards making GC more hackable again, so a future
GCologist will find it easier to implement generational GC

> Would we want/need to move to a generational GC if we had what you
> describe?

Yes. Absolutely.

As I said above, I think this might actually help generational GC
implementations a lot: those are going to want a "tell me which
objects would be collected if I ran mark-and-sweep now" function for
checking, and that's what this is.

Pip