fix: comments above bound-call lines no longer corrupt parsing

[danieljohnmorris]

Summary

An indented -- comment immediately above a paren-bound call silently corrupted parsing. The function body ended up empty and the diagnostic pointed many lines past the actual cause, typically inside a format-string {} placeholder. Manifesto-wise, comments are supposed to be free; this regression made them a 15-minute-per-occurrence tax.

Repro

Before:

$ cat repro.ilo
main>t
  -- build the thing
  m=(fmt "k={}" 1)
  m

$ ilo repro.ilo --run-tree main
{"code":"ILO-T008","message":"return type mismatch: expected t, got _", ...}

After:

$ ilo repro.ilo --run-tree main
k=1

Root cause

normalize_newlines in src/lexer/mod.rs rewrites \n+indent to ; BEFORE the logos lexer's --[^\n]* comment-skip runs. On an indented comment line, the trailing \n is also rewritten to ;, so the logos regex (which stops at \n) greedily eats the comment text plus every following statement up to the next non-indented newline. The body vanishes, the verifier reports a type mismatch with no useful position, and the ILO-P009 expected expression, got Semi cascade points at a { inside the format string several lines later.

What's in the diff

• Commit 1 moves comment-handling into normalize_newlines itself. When it sees --, it advances past the comment content without emitting anything and leaves the trailing \n intact for the loop's existing newline handling. It also passes through "..." string literals verbatim so -- inside a string isn't mistaken for a comment, and suppresses duplicate ; emission when a comment-only line precedes an indented continuation.
• Commit 2 adds tests/regression_comment_parse_corrupt.rs. 15 tests across tree / VM / cranelift covering: paren-bound fmt with leading comment, mid-body comment between bindings, stacked comments, comment containing {}/;/(), and dashes inside a string literal.
• Commit 3 adds examples/comment-above-call.ilo with -- run: / -- out: assertions so tests/examples_engines.rs exercises the same cases at the higher integration layer across every engine.

Test plan

• Original repro from ilo_assessment_feedback.md L1790 returns k=1 instead of ILO-T008
• cargo test --release --features cranelift full suite passes (2866 lib tests, plus all integration tests, including the new 15-test regression_comment_parse_corrupt)
• tests/examples_engines.rs passes with the new example
• cargo fmt --check clean
• cargo clippy --release --features cranelift --all-targets -- -D warnings clean
• -- inside a string literal ("hello -- world") still round-trips correctly

Follow-ups

None. The fix is local to normalize_newlines and matches the logos --[^\n]* rule it's now running ahead of.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!