Skip to content

TODOs.md

Latest commit

 

History

History
35 lines (34 loc) · 2.58 KB

TODOs.md

File metadata and controls

35 lines (34 loc) · 2.58 KB

TODOs

  • CircleCI setup — build, test on every push
  • x86_64 reference counting — 1 test failure remaining (memReclaimBurn). RC infrastructure is built but disabled. See docs/x64-refcounting.md.
  • AI-agent doc audit: review AGENTS.md / README / docs vs. the actual code and dev setup. Specific gaps to close:
    • Write docs/adding-a-backend.md (scattered across PR description, architecture.md, and x64-refcounting.md today)
    • Write docs/features/x64-codegen.md to mirror the existing arm64-codegen.md
    • Audit docs/features/*.md for stale ARM64-only references (binary-generation.md, register-allocation.md especially)
    • Verify docs/adding-features.md covers per-arch pass changes
    • Refresh the AGENTS.md benchmarks section (drop "ignore quicksort error"; point at benchmarks/README.md)
  • Write docs/worktrees.md if the worktree workflow becomes more elaborate.
  • x86_64 encoding test coverage — 36 of 47 instruction types untested at byte level
  • AST.FuncRef — what's the plan for this? Looks dead but idk. Added in 6945c98 for higher-order functions, but the current pipeline routes them through replaceFuncRefsWithWrappers (2_AST_to_ANF.fs:2615) which rewrites AST.Var name → AST.Closure (wrapperName, []). I can't find anywhere AST.FuncRef is actually constructed, yet it's pattern-matched in ~15 sites across 2_AST_to_ANF.fs, 1.5_TypeChecking.fs, ASTPrettyPrinter.fs, IRPrinter.fs, 2.5_RefCountInsertion.fs, 3_ANF_to_MIR.fs, ANFDeadCodeElimination.fs, both parsers (validation only), and E2ETestRunner.fs. Was this left in for a future pass, or is it just leftover from the original design? If dead, removing it would delete ~60-80 lines. (Note: ANF.FuncRef is a different type and is still live — constructed in 2.6_PrintInsertion.fs:50.)
  • Run compiler against existing package repo
  • Additional architectures (arm32?)
  • int64 assumptions
  • add values
  • mutmut testing
  • matching darklang language
  • increasing code coverage
  • completing benchmarks
  • expanding to support full language
  • support full darklang stdlib
  • Json stdlib module (parsing/serialization)
  • support full darklang test suite
  • reimplement darklang compiler in Darklang
  • reimplement test suite in Darklang
  • complete Unicode string support
  • add optimizations
  • remove crashes
  • end-to-end SSA
  • SSA-based HIR (sub ANF?)
  • SCCP-based HIR, MIR, and LIR optimizations
  • remove non-functional idioms
  • unify memory management (RawPtr, heap primitives, reference counting)