But modern CPUs are all converging on 64-bits, for example.

Use of SIMD intrinsics are a tougher nut, but I've actually been playing with them at an IR level for hobby stuff and I declare its not intractable.

Converging on 64-bit CPUs doesn't mean converging on standard `sizeof()` values (c.f. windows x86_64 ABI is LLP64 but linux x86_64 ABI is LP64)

It's easy to loose sight of the fact that we are talking about 64bit OS X here... The diff between ir that targeted x86-64 and arm64 and so on is something Apple has some control over.

I know, I play with a hobby llvm backend that retargets.

true, I didn't mean that it would solve a future 32 v 64 problem.