locale_t etc are unusable

[Hello71]

18:51 <Hello71> how do I check if a libc::locale_t is null?
18:52 <Hello71> it's just an enum, so I can't call is_null on it
18:52 <Hello71> as *const u8 == null? :/
18:53 <talchas> Hello71: you should never have a locale_t
18:53 <talchas> you should only ever have *const locale_t
18:53 <Hello71> then how do I use uselocale and newlocale and so on
18:53 <Hello71> those return locale_t, not *const locale_t
18:54 <Hello71> because C locale_t is really a pointer to a struct. it has to be, because (locale_t)0 is used everywhere
18:56 <talchas> hmm, that would be a bug in the libc crate then
18:56 <talchas> yeah, that's wildly incorrect
18:56 <talchas> having a value of enum Foo {} is UB
18:57 <Hello71> sounds about right
18:57 <Hello71> I don't like how POSIX likes to define x_ts as primitive types sometimes and structs other times and sometimes pointer types
18:58 <occultus> That means uselocale is completely uncalleable from safe code. Is newlocale actually bound?
18:58 <Hello71> actually I guess locale_t can be an integral type, but I'm willing to wager a guess that it's a pointer to struct everywhere
18:59 <talchas> occultus: yeah, it's just a bug in libc, and given that the current definition is 100% worthless it's probably safe to just change its definition of locale_t to something useful
19:00 <occultus> oh wait, newlocale is also uncallable becuase it takes a locale_t for base. heh.
19:02 <occultus> talchas: does it qualify as a bug or just not yet implemented?
19:04 <talchas> occultus: given that it's providing the type and fns, I'd call it a bug
19:13 <Hello71> https://github.com/rust-lang/libc/blob/master/src/fuchsia/mod.rs#L73 lmao
19:14 <Hello71> talchas, occultus: ^
19:14 <occultus> my guess is they just thought "if they're uninhabited nobody can use them, so nothing bad can happen"
19:14 <occultus> except that something bad does happen, people see them in the docs but can't use them
19:15 <talchas> Hello71: well, that's even just weighing in on whether you want *const empty_struct or *const empty_enum
19:16 <talchas> (ie, it probably wants to be struct DIR(()); or equivalent so that accidentally dereferencing a *const DIR is useless rather than UB)

[alexcrichton]

seems reasonable to me to change! It sounds like no one's likely relying on the current implementation, so we can probably tweak it.