Coverage

Where aowlparser stands against native nifler, and what it deliberately does not do.

• Parity
• Byte-exactness
• What nifler catches that aowlparser does not
• Not attempted

---

Parity

target	files	structural	byte-exact
nimony/src	184	184	184
nimony/lib	105	105	91
upstream Nim/lib	310	310	283
curated corpus	172	172	156

599 valid files parse to a tree structure-identical to nifler's, with zero crashes and zero hangs. Every construct the nimony compiler's own source and the full upstream Nim standard library exercise round-trips. Measured by tests/stress.sh.

Byte-exactness

Structural parity means the same token tree; byte-exactness additionally means the same relative line-info on every node. It is complete on nimony/src (184/184) and holds for 558 of the 599 valid files overall. The remainder are structurally identical but differ in some line-info deltas — same tree, same tags, a different @col,line on a few nodes.

Byte-exactness came down to matching nifler's anchoring model (relLineInfo(n, parent), stamping each node with n.info − parent.info) per construct — which source token each node anchors at. The non-obvious anchors:

• Commands — a statement command anchors at the callee's info (the . for a dotted callee); an expression command anchors at the first argument, giving the callee a negative delta.
• Module stmts — anchors at the first real token (a leading ## counts, a plain # does not).
• Name wrapping — Name* is nkPostfix (anchor = *); Name {.p.} is nkPragmaExpr (anchor = {.), for section members and object fields alike.
• do-notation — the call anchors at the callee's .; the do node itself anchors at the body's first token, not the do keyword.
• Portable paths — nifler relativises the recorded path to the cwd by default; mirrored via --portable-paths.

What nifler catches that aowlparser does not

A residue of indentation-context errors — cases nifler rejects by tracking a running indentation stack the range-splitter does not maintain. These are left out on purpose: detecting them risks a false positive on valid code, which would make the tool worse than nifler, not better. Everything aowlparser flags is proven zero-false-positive against the 599 valid files and the full Nim standard library.

For what aowlparser catches beyond nifler — recoverable diagnostics with fix-its and related locations — see Configuration → Diagnostics.

Not attempted

aowlparser is syntactic only. Type and semantic errors (undeclared names, type mismatches, effect violations) surface later, from the semantic pass over the typed .s.aif. Semantic suggestions (e.g. x == 5 used as a statement, && for and) belong to a separate lint layer, not the parser.