Full diff on the test suite:https://gist.github.com/gzm0/b6b7b064f0edadbd1b65af827957d9c9 (fast opt).

@sjrd, a somewhat related question: With nullability in the IR typesystem, why are non-nullable boxed types not subtypes / equivalent to primitive types?

It seems, this appears quite often in the diff:

-    var t$1 = $uJ(new $c_RTLong(lo$3, hi$4));-    var lo$4 = t$1.RTLong__f_lo;-    var hi$5 = t$1.RTLong__f_hi;

That not only removes an unnecessary unbox but also an unnecessary allocation.

@sjrd, a somewhat related question: With nullability in the IR typesystem, why are non-nullable boxed types not subtypes / equivalent to primitive types?

They may have a different representation. A non-nullableCharacter is an instance of$Char. Achar is a number. The phenomenon is exacerbated in Wasm.