← board

NilPy: total = total + k * v with a VARIANT operand fails to lower

Found 2026-07-20 while landing [[feature-nilpy-missing-builtins]]; confirmed PRE-EXISTING on a compiler built from committed HEAD with that work reverted, so it is not part of it.

Repro

xs = [10, 20]
total = 0
for v in xs:
    total = total + 2 * v      # v is a for-in VARIANT
print(total)                   # CPython 60



error: IR_UNSUPPORTED: frontend could not lower AST node (kind 5) — a frontend
gap, would miscompile

Kind 5 is AN_BINOP. Each half works ALONE — print(2 * v) lowers (pymul_v), and total = total + v lowers — so it is the nesting of a variant-producing multiply inside an addition that has no path: the + sees a Variant left and a Variant-typed BINOP right, and the variant-add lowering evidently only handles a leaf operand there.

Loud, not silent, which is why it is a 60 and not urgent. But it blocks the most ordinary accumulation loop there is, and every for-in loop variable is a variant, so it will be hit constantly by real corpus code.

Shape

ir.inc's variant arithmetic (the pymul_v / variant-add arms) should lower a nested variant BINOP operand by materialising it into a variant temp first, rather than requiring a directly-addressable operand. Compare the IRVariantAddr path — the same "a variant RVALUE has no address" constraint that [[project_variant_scalar_unbox_landed]] records.

Gate

test-nilpy green + --tier quick + self-host byte-identical, with the repro plus total = total + v * v, total = total + (v + 1) * 2 and a float variant matching CPython.

Log