search

softdevteam / yksom

Other
8 stars 6 forks source link

Null entries on the SOMStack after they're removed. #206

Closed ltratt closed 3 years ago

ltratt commented 3 years ago

This prevents the GC thinking they're still alive. It would be preferable to tell the GC "only scan the first N words of the stack" but we don't have any way of doing that at the moment AFAIK.

This doesn't have any performance difference that I can measure either way.

jacob-hughes commented 3 years ago

It would be preferable to tell the GC "only scan the first N words of the stack" but we don't have any way of doing that at the moment AFAIK

There is an API for this in Boehm, but I'm not sure how we'd use this with SOMStack. Maybe something like new_from_layout_with_scan_limit(layout: Layout, limit: usize).

These changes LGTM though.

bors r+

ltratt commented 3 years ago

There is an API for this in Boehm, but I'm not sure how we'd use this with SOMStack. Maybe something like new_from_layout_with_scan_limit(layout: Layout, limit: usize).

It's not a limit that you know at creation time: it's dynamic. So we'd have to have some way of either reading a usize or calling a user function to say "tell me how much of this chunk of memory is actually in use".

[FWIW, I am now wondering what happens when you do pop on a Rust Vec... If it doesn't null out the deleted entry -- which it might not do so as there's no correctness reason for it to do so -- we'll get memory leaks.]

bors[bot] commented 3 years ago

Build succeeded: