fix chained nil-coalesce mis-parse on arity-aware calls

[danieljohnmorris]

Summary

Chained mget m "a" ?? mget m "b" ?? 99 failed at the verifier with ILO-T006 arity mismatch: 'mget' expects 2 args, got 3. Single mget m k ?? d was fine; only chains broke.

Root cause is in parse_call_or_atom's post-arg break. After each collected argument, the loop checks for an infix operator to stop on, but allows the prefix-binary lookahead (looks_like_prefix_binary) to override that break. ?? is registered as a prefix-binary candidate (so g ??c 7 can pass a prefix-?? sub-expression to g), and ?? mget m "b" has 3 tail atoms which the scanner happily reads as a prefix nil-coalesce. So the first mget kept absorbing ?? mget m as its third arg, then "b" ?? 99 as its fourth, producing the bogus 3-arg verifier error.

Fix: once at least one argument has been collected for a call, ?? is unambiguously the infix nil-coalesce on the call result. Don't let the prefix-binary scanner pull it back into call-arg territory. The first-token case is left alone so prefix ??c 7 in a call-arg position still works.

Repro

f>n;m=mset (mmap) "a" 1;mget m "a" ?? mget m "b" ?? 99

Before:

ILO-T006 arity mismatch: 'mget' expects 2 args, got 3

After:

1

What's in the diff

• src/parser/mod.rs: in the post-arg infix check in parse_call_or_atom, treat Token::NilCoalesce as always-infix. One condition, one comment block explaining why.
• tests/regression_mget_default.rs: extend the existing cross-engine harness with chained mget m k1 ?? mget m k2 ?? d in first-hit, second-hit, both-miss and no-default shapes. Add an at-based pair to confirm the fix isn't mget-specific. Drop the stale comment that said chained ?? didn't parse.
• examples/chained-nilcoalesce.ilo: in-context learning example exercised across every engine by tests/examples_engines.rs.

Test plan

• Repro program returns the expected value across --run-tree, --run-vm, --run-cranelift.
• Existing regression_prefix_nil_coalesce (covers g ??c 7) still passes — prefix-?? in call-arg position is unaffected.
• Existing regression_mget_default single-mget cases still pass.
• examples/chained-nilcoalesce.ilo runs clean under the examples harness.
• Full suite green: cargo test --release --features cranelift.
• cargo fmt --check, cargo clippy --all-targets --release --features cranelift clean.

Follow-ups

None. The post-arg infix check was the only place this leaked.

[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!