Skip to content
nex

Roadmap

Sixteen phases from lexer to a tagged v0.1.0 release with a package manager, language server, standard library and documentation site. Progress is tracked one micro-step at a time in progress.txt, one step per commit, no skipping ahead.

Current position

Section titled “Current position”
  • last step: 5.11
  • current phase: 5 (tree-walking interpreter)
  • next step: 5.12, the print, str, int and float builtins

Phase status

Section titled “Phase status”
Phase Milestone State
0 Foundations done
1 Lexer done
2 AST design done
3 Parser: expressions done
4 Parser: statements & items done
5 Tree-walking interpreter in progress
6 Name resolution + type checker planned
7 Language feature wave 2 planned
8 LLVM backend planned
9 WASM target planned
10 Standard library planned
11 Package manager planned
12 Editor highlighting planned
13 Language server planned
14 Documentation site planned
15 Polish + v0.1.0 planned

What each phase delivers

Section titled “What each phase delivers”

0, Foundations. Workspace, CI, examples, progress tracking.

1, Lexer. The full token stream, with error recovery. Done; see the lexical structure reference.

2, AST design. Node identity, spans, and the node types for expressions, statements, items, types and patterns, plus an s-expression pretty printer for snapshot tests. Done; AST coverage lists what the tree still can’t represent.

3 and 4, Parser. Expressions first, then statements and items. Error recovery and ariadne diagnostics land at the end of phase 3.

5, Tree-walking interpreter. nex run starts working.

6, Type checker. Name resolution and static typing. This is where v0.1 of the language becomes real.

7, Feature wave 2. Block comments, character literals and the other missing pieces. The spec freezes as v0.1 at the end of this phase.

8 and 9, Backends. LLVM native codegen for nex build, then WebAssembly via --target wasm32. Every backend has to produce identical output on the regression corpus.

10 and 11, Standard library and package manager. nex test arrives alongside the stdlib and its test runner.

12 to 14, Tooling. Editor highlighting, the language server, and this documentation site.

15, Polish. The nex fmt formatter, benchmarks, and the v0.1.0 tag.

Deferred / pending

Section titled “Deferred / pending”
  • CI doesn’t install LLVM yet. Needed from step 8.2.
  • Parser throughput on a 10k-line file: ~274k lines/sec debug, ~948k release. Re-measured by crates/nex-syntax/tests/throughput.rs.
  • Block comments (/* */) arrive in Phase 7; they currently lex as operators.
  • Character literals ('a') are not in the language.

Design notes that shape later phases

Section titled “Design notes that shape later phases”
  • NodeIds are dense and sequential so they can index side tables directly. Later passes store their results in side tables keyed by NodeId rather than mutating the AST, which stays immutable after parsing.
  • NodeId::DUMMY marks nodes invented during error recovery. It panics on .index() so a dummy can’t silently corrupt a side table.
  • Memory management is decided in Phase 8. Current plan: automatic reference counting emitted by the code generator for heap values (str, arrays, boxed enums). No ownership or borrow checking in v0.1.