changes after review

* add ValueType::cell_cast which computes the canonical result type
  when applying a cast operator.  For now, only allow double as scalar.
* slight change to make_instruction; add the cell type as an extra
  parameter and require consistency (future-proofing).
* In reference implementation, copy values via a temporary of
  the appropriate type, ensuring that casts may lose precision
  like they should.

1 files changed, 2 insertions, 2 deletions

diff --git a/eval/src/tests/instruction/generic_cell_cast/generic_cell_cast_test.cpp b/eval/src/tests/instruction/generic_cell_cast/generic_cell_cast_test.cpp
index 26dd9737fe1..09596301fab 100644
--- a/eval/src/tests/instruction/generic_cell_cast/generic_cell_cast_test.cpp
+++ b/eval/src/tests/instruction/generic_cell_cast/generic_cell_cast_test.cpp
@@ -37,8 +37,8 @@ TensorSpec perform_generic_cell_cast(const TensorSpec &a, CellType to, const Val
 {
     Stash stash;
     auto lhs = value_from_spec(a, factory);
-    auto res_type = ValueType::make_type(to, lhs->type().dimensions());
-    auto my_op = GenericCellCast::make_instruction(res_type, lhs->type(), stash);
+    auto res_type = ValueType::cell_cast(lhs->type(), to);
+    auto my_op = GenericCellCast::make_instruction(res_type, lhs->type(), res_type.cell_type(), stash);
     InterpretedFunction::EvalSingle single(factory, my_op);
     return spec_from_value(single.eval(std::vector<Value::CREF>({*lhs})));
 }