128-bit integers marked as having an unknown stable ABI

[haraldh]

Introduced by 549dd10 128-bit integers used in e.g. extern "sysv64" produce the warning:

warning: `extern` fn uses type `u128`, which is not FFI-safe
   --> src/xxx.rs:126:6
    |
126 | ) -> u128 {
    |      ^^^^ not FFI-safe
    |
    = note: `#[warn(improper_ctypes_definitions)]` on by default
    = note: 128-bit integers don't currently have a known stable ABI

warning: 1 warning emitted

I dispute this statement by pointing to the now de facto source of the x86-64 System V ABI:

• https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/low-level-sys-info.tex#L552-560
• https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/low-level-sys-info.tex#L767-768

[est31]

Good point actually. Allowing it for extern "C" for sysv64 using targets would probably be a bad idea due to the current lack of a portability lint (#41619), but allowing it for extern "sysv64" should be doable I think.

cc @nagisa

[est31]

Made a PR to allow it on sysv64: #78481

[workingjubilee]

Making some notes here for the next swing at a generic integers RFC:

• _BitInt(N) is upcoming in C23
• Current spec for SysV x86-64 _BitInt(N) where N > 64 is as-if eightbyte-aligned structs
• Thus, __int128 and _BitInt(128) are ABI-incompatible
• If we RFC something like int<const N> we may want to decide:
  • type i128 = int<128>;
  • int<N> is repr(C)-compatible as _BitInt(N)
• That would make this pair incorrect:

extern "C" fn by_value_int128s(a: i128, b: i128, c: i128, d: i64, e: i128) -> i128

__int128 by_value_int128s(__int128 a, __int128 b, __int128 c, int64_t d, __int128 e);

• Deciding int<128> =! _BitInt(128) means we can't call this C23 function:

_BitInt(128) by_value_big_bitints(_BitInt(128) a, _BitInt(128) b, _BitInt(128) c, int64_t d, _BitInt(128) e);

I have opened an issue regarding this at the x86-64-ABI gitlab. However they may have good reasons for their current spec, so it is not guaranteed they will consider this a spec bug.

[tgross35]

I think the links in the original issue are dead, unless this was actually the intended highlight range

(GL changes the URL if you hit the "permalink" button top right, not like GH where you can copy a permalink directly)

[tgross35]

Did a bit of sleuthing, I believe these are the two relevant sections (based on the commit at time of OP):

https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/ca48acd297294172349d4057fb0f04016c5733bd/x86-64-ABI/low-level-sys-info.tex#L591-602

https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/ca48acd297294172349d4057fb0f04016c5733bd/x86-64-ABI/low-level-sys-info.tex#L814-815