Unboxed Int32.t on 64-bit?

[mbacarella]

@vlaviron

If you have time to spare, you might want to check the ocplib-endian library (it’s on opam). It exposes compiler primitives to read and write numbers from raw strings (and bytes).

What’s the advantage to reading and writing strings vs (integer) arrays? @copy also suggested this, although in my mental model they should be the same thing?

[dbuenzli]

@vlaviron

If you have time to spare, you might want to check the ocplib-endian library (it’s on opam). It exposes compiler primitives to read and write numbers from raw strings (and bytes).

Note that since 4.08. These are also exposed by the Stdlib. See https://github.com/ocaml/ocaml/pull/1864

[vlaviron]

@mbacarella

What’s the advantage to reading and writing strings vs (integer) arrays? @copy also suggested this, although in my mental model they should be the same thing?

It takes less memory; to represent an array of int32 values with n elements, an int array takes 8 * n + 8 bytes, while a string/bytes takes 4 * n + 16 bytes (or 4 * n + 12 if n is odd).
It should also be very slightly faster to use operations on unboxed int32 than regular int, because the latter is represented with the least significant bit set to 1 and this has to be preserved.

@dbuenzli

Note that since 4.08. These are also exposed by the Stdlib . See Functions to read/write binary representations of numbers by alainfrisch · Pull Request #1864 · ocaml/ocaml · GitHub

I had forgotten about that, thanks for reminding me. Although ocplib-endian still exposes more functions than the ones in the Stdlib (unsafe versions, String and Bigstring versions).