starlight VS pika

I have great news

Hello to all! This crate implements Shadow Stack algorithm which allows to track GC objects on stack with almost zero overhead! This algorithm is used inside Restricted Python and seems to work very well. This crate soon will replace DIY shadow stack implementation in starlight(JS engine in Rust) too.

Working on startup snapshots in starlight. I already have very basic implementation which allows to initialize runtime in just 17 microseconds from snapshot vs 23 microseconds without snapshot when every builtin is created from scratch. Future work is aimed mostly on making deserialization faster

Hi to all! I'm working on starlight (JS engine in Rust) and I can't choose memory management technique: Right now I have conservative on stack precise on heap GC which somehow manages to work but still has segfaults and I'm also working on rcgc feature which will use RC as GC algorithm but my main question: is it worth using RC over tracing cycle and how hard it will be to implement JIT when RC is used? I've never seen any runtimes that use RC and implement JIT.

There's test262_passed file in repo, you can take a look at what tests pass :)

I just switched Pika to using significant indentation. This is mostly because of how annoying line continuation is in a ML-style language (so f a b syntax) without significant indentation or required semicolons, but you can read more about the reasons for that decision in this section of the README.

Recently, I've been working on adding garbage collection to Pika. I've successfully written an Immix-based garbage collector that works with the LLVM GC support infrastructure, and I'm currently working on integrating the GC with Pika, or really Durin, the dependently-typed intermediate representation that Pika compiles to. Because types are passed around at runtime, objects of unknown type and size can be stored unboxed in polymorphic data structures; but that makes keeping track of type information for heap allocations somewhat harder, because type information needs to be allocated and constructed at runtime in some cases. It's an interesting design problem, because you want constructing type information to be fast; but the GC will run much more often, so maximizing tracing speed by avoiding e.g. indirection in type information is important; and you also want to construct as much type information as possible at compile time and embed it as constants.

I've been working on adding algebraic effects to Pika during the past month. It's turned out to be harder than I thought it would, but I'm almost done - performing and catching one effect at a time works, and the compilation strategy I'm using now (I reimplemented the whole thing after realizing the strategy I was using wouldn't actually work) should be enough to handle multiple effects at once and also effect polymorphism, I just have to get those working in the elaborator.

After taking a break from Pika, my dependently-typed ML for systems programming, during the month of January, I've started working on it again by getting recursion to work properly. I'm planning on starting to implement algebraic effects next, and an Immix-based garbage collector for boxed values after that. Here's what my current plan for algebraic effects looks like: