NilPy: a bitwise op on a FLOAT variant truncates instead of raising TypeError

Found 2026-07-20 while adding variant bitwise operators for the uforth drive.

EmitVarBinOp (ir_codegen.inc) lowers variant arithmetic in two paths: an integer path and a double path, chosen by the runtime tag. The new bitwise operators (& | ^ << >>) are integer operators, so on the DOUBLE path they truncate both operands to Int64 and apply the op, producing VT_INT64.

Python raises TypeError: unsupported operand type(s) for &: 'float' and 'int'.

Why it shipped this way

Emitting a runtime type error needs machinery EmitVarBinOp does not have (it is hand-emitted x86-64 with no error-raise helper in scope). Truncation is at least consistent with how div/mod ALREADY treat a double operand in the same routine — those deviate from Python too, deliberately, with Pascal semantics.

Unreachable from the uforth corpus, which masks integers behind isinstance guards, so this is a latent correctness gap rather than an active wrong answer.

Fix when picked up

Give the variant-arithmetic path a way to raise, then make float-operand bitwise raise TypeError. That same helper would let div/mod-by-zero and other variant type errors stop deviating as well, so it is worth more than this one bug.

Note

x86-64 only: EmitVarBinOp has no counterpart in the other backends, so variant bitwise ops are not available cross-target at all yet.

Log

• 2026-07-23 — resolved, commit 64d86ef3.