Result: done
Files: crates/nikaia-std/src/tools/check_core_args.nika crates/nikaia-std/src/tools/check_core_leaves.nika crates/nikaia-std/src/tools/check_core_stmt.nika crates/nikaia-std/src/tools/check_facts.nika crates/nikaia-std/src/tools/emit_core_flow.nika crates/nikaia-std/src/tools/lowered/check_core_args.rs crates/nikaia-std/src/tools/lowered/check_core_leaves.rs crates/nikaia-std/src/tools/lowered/check_core_stmt.rs crates/nikaia-std/src/tools/lowered/check_facts.rs crates/nikaia-std/src/tools/lowered/emit_core_flow.rs tests/compiler/tests/a_let_lambda_handed_where_a_function_is_kept.nika tests/compiler/tests/a_let_lambda_handed_where_a_function_is_kept.stdout
Commit: check: a let-bound lambda handed where a function is kept is kept (Fixes #1043)
Body: Option (i). An untyped let that binds a lambda records it in let_lambdas, keyed by the binding. Where that name is kept (a parameter the callee keeps, a returned fn, a field), check_core_a_kept_lambda keeps the let’s lambda once, as it does for a typed let, marks it in let_kept_lambdas and takes the name out of run_code. The #997 look-ahead now types the lambda from a keeping callee too. Without a declared type rustc cannot unsize Kept<closure> on its own, so the emitter writes the parameters with their types by use and casts the Arc to dyn Fn(_..) -> _, adding + Send + Sync where the host crosses into user code.
Fixes: #1043
Tests: tests/compiler –both-settings 434 passed, 1 failed (a_trait_column_change_warns, settings disagree on cargo file-lock lines, not this change); new test and the 3 model tests ok; it guard set + selfhosting_defects:: 43 passed (sequences guard passed; it is on Red-on-develop at 18/16); clippy clean; unfixed: rows a and b failed (rustc mismatched types) on the pre-fix binary as probes, not by stash (forbidden in the shared checkout); where.nika lists no tests for check_core_a_kept_lambda
Gate: green
Deviation: none
Found: (c) a kept fn field read through a struct, let h = H { f: g }; h.f(1) with g from an untyped let, is refused at check with NK2605 “H.f can fail with ?” whether or not this fix is in (a separate cause, so row c is not in the test). (d) #997 is incomplete: let add = fn(x) { x + 3 }; apply(add), where apply(f: fn(i64) -> i64) only runs f, gets past check and then rustc refuses it with “i64 is not a future”: the closure is handed as &add where &impl AsyncFn is wanted. It fails the same way on the unfixed compiler, and this fix leaves its lowering unchanged.
Needs: none
Changelog: A lambda named by an untyped let and handed where a function value is kept is kept, instead of a lowering rustc refused.
Owner: none
Disk freed: 0 GB (cache 3.3G kept, 7.5G free on /)
Rs2nika: none
Pushed: 599cef06 (pushed by drive.nika)
