Bytecode VM, dynamic number package, and R7RS compliant Scheme implementation.
Ava Affine
cf626b2bfc
All checks were successful
per-push tests / build (push) Successful in 36s
per-push tests / test-utility (push) Successful in 36s
per-push tests / test-frontend (push) Successful in 37s
per-push tests / test-backend (push) Successful in 31s
per-push tests / timed-decomposer-parse (push) Successful in 34s
This commit removes the dependency upon Mycelium::sexpr::Datum from Hyphae
and instead adds a heap.rs module including three types:
- Gc: This is essentially a Box<Rc<T>> but passed around as a *const Rc<T>.
not only does this allow the wrapped T to be passed around in a format
that fits completely within a single physical register, this also
applies a greedy reference counting garbage collection to each and
every object allocated in the VM.
- Datum: This is a simplified enum for type polymorphism. Similar to the
original Mycelium::sexpr::Datum.
- Cons: This is a very standard Cons cell type.
Additionally, two new instructions are added:
- DUPL: a deep copy instruction
- CONST: an instruction to create number datum from embedded constants
- NTOI: casts a number to its inexact form
- NTOE: casts a number to its exact form
Fixes: #35 and #36
Signed-off-by: Ava Affine <ava@sunnypup.io>
|
||
|---|---|---|
| .forgejo/workflows | ||
| decomposer | ||
| hyphae | ||
| mycelium | ||
| organelle | ||
| snippets | ||
| .gitignore | ||
| .gitlab-ci.yml | ||
| Cargo.lock | ||
| Cargo.toml | ||
| LICENSE | ||
| readme.md | ||
Mycelium
Mycelium aims to provide an efficient compiled to intermediate bytecode implementation of R7RS Scheme as well as a virtual machine that will execute such bytecode.
Two major use cases are at the forefront of mind when designing and implementing this project: a POSIX shell interpreter as well as a compiled to bytecode language for running on ESP32 devices.
Current Status
Currently the lexer and parser are implemented. On an X86 machine equipped with 64GB RAM and an AMD Ryzen 7900 CPU this lexer and parser are capable of creating a fully validated abstract syntax tree from approximately 11200 lines of handwritten scheme in about 55 milliseconds on average.
Currently the bytecode VM and its instruction set are next to implement.