Conversation

Patrick Walton
@pcwalton
My strong feeling is that the "memory safe systems programming language that's easier to use than Rust" that everyone wants is going to have to have something that looks a lot like a garbage collector. There's a lot of room to design a fast GC for "systems-like" patterns.
35
369
Patrick Walton
@pcwalton
You could imagine a system that integrated tracing GC for language-native objects with reference counting, or conservatively scanned C heaps, for FFI objects, for instance.
6
65
James Tucker
@raggi
Replying to
@DanielMicay
and
@pcwalton
so the standard we could do with is a standard marker in a thread local for "don't migrate this stack to somewhere else", with some push to get common libraries to implement the guards.
Daniel Micay
@DanielMicay
Replying to
@DanielMicay
and
@pcwalton
nginx has a variant of JS (njs) that's scoped to web server requests so there's mostly no use for doing GC. You can use arrays, objects, closures, etc. It allocates with the usual nginx thread-local allocation pools and then all of the JS memory is freed when the request ends.
3
Bakerposting
@bakerposting
Replying to
@DanielMicay
and
@pcwalton
Barriers are quite cheap, and rarely synchronise (when you run out of thread-local queue in the slow path e.g.) dl.acm.org/doi/10.1145/24 It's more tracing that gets you; thread-local GC is a worthwhile idea - fitting a nursery in cache would make scanning it fast.
dl.acm.org
Barriers reconsidered, friendlier still! | ACM SIGPLAN Notices
Read and write barriers mediate access to the heap allowing the collector to control and monitor mutator actions. For this reason, barriers are a powerful tool in the design of any heap management...