fix: suppress top-level auto-print when entry body ends in a loop

[danieljohnmorris]

Summary

Print-loops like @x xs{prnt x} at the function tail used to double-print at the top level: each item from the loop body via prnt, then the loop's last-body-value bubbled up as the program result and got auto-printed again. The well-known persona workaround was a trailing +0 0 sentinel after every print-loop, called out as a "consistent irritant across all 5 programs" in the assessment doc (line 834) and as one of the top friction items at line 854.

Print-layer-only fix, scoped to src/main.rs. When the program's syntactic entry-function body ends with @/wh AND has no early-return path, the plain-text top-level auto-print is suppressed. Loop-as-expression semantics inside functions are completely unchanged - a caller that consumes such a function's return value still sees the loop's last body value. JSON mode is untouched so machine-readable consumers always get a structured result.

The check is syntactic, not value-based: a function that explicitly returns nil from a non-loop tail (e.g. nothing>O n;nil) still prints "nil" because the user asked for it.

Repro

Before:

$ cat /tmp/p.ilo
print-each>n;xs=[1 2 3];@x xs{prnt x}
$ ilo /tmp/p.ilo print-each
1
2
3
3       <-- the loop's last-body-value, auto-printed again

After:

$ ilo /tmp/p.ilo print-each
1
2
3

Cross-engine: tree, VM, and Cranelift all agree.

What's in the diff

Three commits, each one logical step:

1. suppress top-level auto-print when entry body ends in a loop - the helper (program_result_should_suppress) walks the entry function's AST tail and decides whether to suppress; print_value gains a suppress_loop_tail flag, plain mode honours it, JSON mode ignores it. All four engine runners (run_default, run_cranelift_engine, run_vm_with_provider, run_interp_with_provider) compute the flag once at dispatch and thread it to print_value. A separate body_has_early_return walker is the conservative safety valve: if the body contains any ret or braceless guard anywhere (including inside nested guards/match/loops), suppression is off, because we can't tell at print time whether the program's value came from the loop tail or from an explicit return.

2. update bare-loop-tail tests and example to use sentinel form - three pre-existing tests asserted the previous "function ends with a bare loop, auto-print the loop value" shape. They're updated to bind the loop value and end with a trailing expression, the same pattern examples/loops.ilo uses for every existing loop example. Inline comments explain the convention so future readers don't think the rewrite was cosmetic.

3. cross-engine regression test and example for print-loop suppression - tests/regression_loop_print.rs runs nine scenarios across tree/VM/Cranelift: top-level print-loop, nested print-loop (must still return the value), empty-loop, brk-loop, while-loop, trailing-sentinel non-regression, early-return + loop tail (conservative rule), explicit-nil return (non-regression), and JSON mode (non-regression). examples/print-loop.ilo documents the same scenarios with -- run:/-- out: annotations so tests/examples_engines.rs exercises them as higher-level engine-parity checks.

Test plan

• cargo fmt --all -- --check clean
• cargo clippy --workspace --all-targets -- -D warnings clean
• cargo test --release --features cranelift - full suite passes
• New tests/regression_loop_print.rs passes on all three engines
• tests/examples_engines.rs picks up the new examples/print-loop.ilo and runs it on tree + VM
• Manual repro confirmed: 1\n2\n3\n instead of 1\n2\n3\n3\n across all three engines
• JSON mode unchanged: --json still emits {"ok":3} for the print-loop case and {"ok":null} for empty-loop
• Nested case verified: outer>n;v=inner();+v 0 where inner ends in a print-loop still returns 3 to outer

Follow-ups

None blocking. If we ever want to revisit the early-return-present-plus-loop-tail edge case, the right place is runtime-flagging "did this value come from the loop tail?" inside the interpreter/VM/Cranelift - significant cross-backend plumbing, deferred unless a persona report surfaces the pain.

[codecov]

Codecov Report

❌ Patch coverage is 94.66667% with 4 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/main.rs	94.66%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!