From what I understand, it risks very large code sizes; after all it is pasting the same code over and over, even if SIMD helps do it in chunks. Loops over tuples up to length 31 are conventionally unrolled with inlined recursive function calls on the Base.tail subtuple, and beyond that you run into type-unstable allocating code. Like I said in a recent related thread, it would actually be cool if we could fill a tuple in a loop that isn’t unrolled (each iteration is like incomplete initialization, but it is complete over the whole loop); maybe this can be done in a special macro or with compiler optimizations over Accessors.jl.