JIT nil-sweep batch 4: text + format helpers

[danieljohnmorris]

Summary

Batch 4 of the JIT-helper permissive-nil sweep, picking up Group B from where #282 (batch 3) left off: the text helpers and fmt2.

Twelve helpers (jit_fmt2, jit_trm, jit_upr, jit_lwr, jit_cap, jit_padl, jit_padr, jit_ord, jit_chr, jit_chars, jit_unq, jit_frq) all silently returned TAG_NIL on type-error, empty-string, invalid-codepoint, and non-finite-element paths where tree and VM both raise an ILO-R009 runtime error. Each helper now carries a packed span_bits immediate and routes its failure paths through jit_set_runtime_error_with_span (the TLS primitive from #254), so the JIT raises the same diagnostic as tree/VM with a caret matching the source span.

(jit_len / jit_str / jit_num from the original batch-4 plan were already routed in batch 3 and intentionally not in scope here.)

Repro before/after

$ ilo 'f>t;trm 1' --run-cranelift f   # before: prints "nil", exits 0
$ ilo 'f>t;trm 1' --run-cranelift f   # after:  ILO-R009 trm requires text, exits 1

$ ilo 'f>n;ord ""' --run-cranelift f  # before: prints "nil", exits 0
$ ilo 'f>n;ord ""' --run-cranelift f  # after:  ILO-R009 ord requires a non-empty string, exits 1

What's in the diff

1. route text + format JIT helpers through JIT_RUNTIME_ERROR — helper bodies in src/vm/mod.rs grow a span_bits parameter and route every error path that previously fell through to TAG_NIL via jit_set_runtime_error_with_span. Both helper registries (the AOT one in compile_cranelift.rs and the in-process JIT one in jit_cranelift.rs) bump the declared arity and pack the source span at every call site. The second registry is the load-bearing bit: without it the JIT would call the helper with one fewer arg and read garbage from an uninit register on the error path.

2. add cross-engine regression suite for batch-4 helpers — new tests/regression_jit_nil_sweep_batch4.rs pins happy-path output of every batch-4 helper across tree, VM, and Cranelift. Per-helper error-path coverage lives next to the helpers as unit tests in src/vm/mod.rs (driving them directly, since the verifier rejects mixed-type programs that would surface these errors via the CLI). tests/regression_ord_chr.rs is consolidated: the previous test that pinned silent-nil-on-empty for Cranelift while tree/VM errored is folded into a single ord_empty_string_errors_on_every_engine now that all three engines agree.

3. add example pinning batch-4 helper happy paths through examples_engines — examples/text-helpers-jit-parity.ilo exercises every helper through the existing tests/examples_engines.rs harness, which runs each -- run: / -- out: pair through every available engine. Doubles as in-context learning material for future agents.

Test plan

• cargo build --release --features cranelift
• cargo test --release --features cranelift — 123 test binaries, 0 failures
• cargo fmt --all -- --check
• cargo clippy --release --features cranelift --all-targets -- -D warnings
• New cross-engine regression suite (14 tests) passes on tree, VM, and Cranelift
• New unit tests for each helper's error paths (22 tests) pass
• examples/text-helpers-jit-parity.ilo runs identically on every engine via examples_engines.rs

Follow-ups

• Helpers in remaining batches (5: collections, 6 already landed via route stats + linalg JIT helpers through JIT_RUNTIME_ERROR (batch 6) #285, 7: I/O) still in flight by sibling agents.
• jit_fmt2 digits-clamping branches (non-finite / negative / >20) stay correct-by-design: they coerce to 0 / 20 to mirror tree, not error.

[codecov]

Codecov Report

❌ Patch coverage is 79.81651% with 66 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/vm/mod.rs	83.11%	39 Missing ⚠️
src/vm/compile_cranelift.rs	43.75%	27 Missing ⚠️

📢 Thoughts on this report? Let us know!