Appearance
Parity — byte-for-byte, from scratch
The goal is blunt: a from-scratch reimplementation of the entire Nim / Nimony toolchain whose output is byte-for-byte identical to the originals — the same parse tree, the same typed IR, the same generated code, down to the byte. Not a fork, not a wrapper around Araq's binaries: each stage is rewritten and then held to the real compiler's exact output by a differential harness.
We are not all the way there yet — and that's the point of this page. It's the honest scoreboard.
Why byte-for-byte
Anyone can write a compiler that's "close." Byte-for-byte is a different bar: it means every naming decision, every emission order, every whitespace convention, every line-info offset matches the reference exactly. It's a brutal, unambiguous oracle — the diff is either empty or it isn't. Hitting it proves the reimplementation didn't just approximate Nim/Nimony's behaviour; it reproduced it. That's what makes each stage a genuine drop-in beside nimony's own.
Where each stage stands
| Stage | ours | vs. | from scratch | byte parity |
|---|---|---|---|---|
| parse | aowlparser | nifler | ✅ | byte-exact on the whole compiler tree (184/184); 91–283 files byte-exact on the stdlib, 100% structural everywhere |
| semcheck | aowlsem | nimsem | ✅ | construct-by-construct; the 166-module corpus is diff-green, the byte-level diff is closing |
| lower | aowlhexer | hexer | ⏳ | currently runs Araq's own 25 passes, so its .c.aif is identical by construction; a from-scratch rewrite is the target |
| C codegen | aowlc | lengc | ✅ | end-to-end correct today (runs, ASan-clean); text byte-parity with lengc is the active push |
| interpret / VM | aowli | (new) | ✅ | two independent engines, byte-identical to each other, held honest against native nimony execution |
| emit → TS / Py / JS / WASM | aowlts · aowlpy · aowljs · aowlweb | (nimony backends) | ✅ | idiomatic, readable output; behaviour-verified against native, run-for-run |
✅ = written from scratch · ⏳ = reuses the reference implementation for now
The parser — proof the bar is reachable
The front of the pipeline is already there. aowlparser is diff-tested against native nifler — structural = token trees equal with line-info stripped; byte-exact = identical .p.aif including every line-info offset.
| corpus | files | structural | byte-exact |
|---|---|---|---|
| nimony/src (the compiler itself) | 184 | 184 | 184 |
| nimony/lib (stdlib) | 105 | 105 | 91 |
| upstream Nim/lib | 310 | 310 | 283 |
| curated | 172 | 172 | 156 |
0 crashes, 0 hangs across all four. The remaining byte gaps are catalogued, not mysterious — see Parity & gaps.
How we measure
Every claim here comes from a differential harness, not a vibe: run the input through both the reference tool and ours, normalize nothing that matters, and diff the bytes. A green diff is the only thing that counts as "done" for a construct; everything else is a punch-list item. That's how the parser got to byte-exact, and it's how each remaining stage closes the gap.