next: issue with domain inference with concat_where at domain boundaries #2205
Description
testee_program(a, b, out, edges, levels) {
__tmp_1 = temporary(domain=u⟨ Edgeₕ: [0, edges[, KDimᵥ: [0, 11[ ⟩, dtype=int32);
__tmp_1 @ u⟨ Edgeₕ: [0, edges[, KDimᵥ: [0, 11[ ⟩
← as_fieldop(λ(a, b) → ·⟪E2Vₒ, 0ₒ⟫(a) + ·b, u⟨ Edgeₕ: [0, edges[, KDimᵥ: [0, 11[ ⟩)(a, b);
out @ u⟨ Edgeₕ: [0, edges[, KDimᵥ: [0, 10[ ⟩
← as_fieldop(
λ(__ct_flat_el_0___tcw_pos, tmp0ᐞ0) →
(λ(_cs_1, _icdlv_6) →
if True ∧ _cs_1 < 9 then _icdlv_6() else if 10 <= _cs_1 ∧ True then _icdlv_6() else ·tmp0ᐞ0)(
·__ct_flat_el_0___tcw_pos, λ() → ·tmp0ᐞ0 - ·⟪Koffₒ, 1ₒ⟫(tmp0ᐞ0)
), u⟨ Edgeₕ: [0, edges[, KDimᵥ: [0, 10[ ⟩
)(index(KDimᵥ), __tmp_1);
}
In this example __tmp_1 is only consumed in KDim: [0, 10[, but computed in [0, 11[. The reason is that the domain inference takes both branches of the if (from concat_where) into account and therefore concludes that K domain needs to have the extra level.