Skip to content

Latest commit

 

History

History
78 lines (60 loc) · 3.11 KB

TODO.md

File metadata and controls

78 lines (60 loc) · 3.11 KB
Raw

TODO

Wish list. Things that could be improved or added.

Parsing, analysis and compilation

  • String literals. See co1.
    • Scan & parse string literals
    • Scan & parse multiline string literals
  • Scan & parse floating point literals. See co1
  • Scan & parse character literals (what type is a char? u8? "rune"?). See co1
  • Arrays and slices
    • Parsing array and slice types e.g. [3][4]type, []type
    • Parsing & analysis of array indexing e.g. a[3]
    • IR gen of array indexing e.g. a[3]
    • Parsing & analysis of array slicing e.g. a[:3]
    • IR gen of array slicing e.g. a[:3]
  • Rename basic primitive types like int32 to LLVM/Rust/Zig style i32, u16 etc.
  • Multi function dispatch e.g. foo(int) vs foo(float32).

IR

Use Co IR in between AST and LLVM IR. This makes it easier to generate code with alternative backends like JS, makes the LLVM backend simpler and removes language-specifics from most of the gnarly codegen stuff.

Keep it (much) simpler than Co1 IR, without regalloc or instruction selection.

  • Dedicated type system that does not rely on AST nodes (IRType)
  • New LLVM backend that uses IR instead of AST as the source
  • Interpreter for testing and comptime evaluation

Testing & QA

  • Parser test infra
    • Scanner/tokenizer tests (src/co/parse/scan_test.c)
    • compile-time AST validator for development & debugging
      • @unres unresolved hierarchy integrity
    • AST repr that outputs easily-parsable S-expr
    • S-expr parser for parsing #*!AST ...*# comments in test sources
    • Diff the expected AST with the actual one. Complain on mismatch. Maybe just text_diff(print(parse(actual)), print(parse(expected))). See co1 which may not be worth porting. Its ergonimics are not great.
  • Codegen tests. I.e. build & run programs and verify their output, like Go examples.
  • GitHub CI with builds See lobster for example: https://github.com/aardappel/lobster/blob/ d8e2ce7f6ce2dd5b94e9bff92532c2e50c438582/.github/workflows/build.yml

Misc

  • Actual full build of packages (need to expand the details of this here)
  • Packages/modules
    • Resolution and importing
    • What are the cached artifacts? Objects? Object archives? LLVM BC? LLVM IR? Nothing..?
  • Comptime eval — a simple interpreter that can evaluate code at compile time. There's some very basic code for array size eval in src/co/parse/eval.c.
  • Explore building WASM for web browsers (without LLVM)
    • Code generation without LLVM to WASM or JS (is there 3rd part stuff I can use?)
  • Look into using mimalloc for memory allocation

Documentation

  • Uhm... documentation?!
  • Language
    • Syntax
    • Semantics
    • Examples