RED-first audit trail (debug skill; fix follows as a separate GREEN
commit). Test-only, +97 lines in the linearity in-source test module,
no production change.
Bug: the [over-strict-mode] lint emits a conservative false-positive.
The consume-detection (any_sub_binder_consumed_for /
pattern_has_consumed_heap_binder, crates/ailang-check/src/linearity.rs
:853,942) only recognises a consume of an (own (con T)) param when a
heap-typed pattern-binder is moved out of `match p`. When p is
destructured into purely primitive fields (e.g. Float/Int) that are
fed into a Term::Ctor rebuilding p's own ctor, that rebuild's
re-consumption of the dismantled allocation is invisible, so the lint
wrongly tells the author `(borrow ...)` would suffice. It is
over-strict, never under-strict (exit 0; codegen/ABI unaffected) — but
an LLM author "fixing" the spurious warning by flipping an export's
declared mode own->borrow would silently invert the ABI ownership
contract, which is why a low-severity advisory FP gets a real
RED-first fix.
The carrier's initial hypothesis (nested inner `match` is the
discriminator) was disproved during diagnosis: the M3
embed_backtest_step_record.ail is silent only because its
implicit-mode scalar Float param disables the lint via the activation
gate (linearity.rs:327), not because it handles the rebuild; the
defect reproduces with a single param and no nesting. The RED unit is
therefore the general synthetic shape — an (own (con T)) param
consumed by a term-ctor rebuild of its own ctor — not the embed
fixtures.
RED test (fails now, goes GREEN when consume-detection sees the
ctor-rebuild re-consumption):
crates/ailang-check/src/linearity.rs ::
linearity::tests::over_strict_mode_silent_when_ctor_rebuilt_from_primitive_fields
Brummel
a11cb7cc9f