add tmpnam and pthread_exit
tmpnam and readdir are trivial IMO.
About the `pthread_create` change: It needs `unsafe` for passing `C` functions to pthread_create (with rust functions the omission of `unsafe` is working of course).
`bindgen` produces this function definition:
```
pub fn pthread_create(arg1: *mut pthread_t,
arg2: *const pthread_attr_t,
arg3: Option<unsafe extern "C" fn(arg1: *mut c_void) -> *mut c_void>,
arg4: *mut c_void) -> c_int;
```
So it would add an additional `Option` around the function. But that would break existing code which uses `libc::pthread_create` and what use is it to call pthread_create without any function pointer, so I left `Option` out.
For reference: I also opened a [stackoverflow question](http://stackoverflow.com/questions/42284562) where the answers were also suggesting adding `unsafe` to the function definition.
Make readdir available on all unix targets
The readdir_r call has problems, and we'll probably want to move to
readdir on many, if not most, unix targets. This patch makes readdir
available in unix, rather than just solaris as before.
See https://github.com/rust-lang/rust/issues/40021
The readdir_r call has problems, and we'll probably want to move to
readdir on many, if not most, unix targets. This patch makes readdir
available in unix, rather than just solaris as before.
See https://github.com/rust-lang/rust/issues/40021
Having the B* constants in `unix/notbsd/mod.rs` passed CI tests
except for powerpc. So we'll try moving into individual
arch/ABI that the CI tests cover for now. This commit should
pass for the following:
- mips32
- mips64
- musl32
- musl64
- android32
- android64
- arm32
- aarch64
- x86
- x86_64
Then we can figure out the powerpc variants. This also prevents
potential errors for sparc64 which is not covered by CI.
Added *_setpshared and *_getpshared bindings
Adding bindings to posix pthreads functions vital for IPC via shared memory.
That's my first PR into libc and I'm not proficient in unix systems programming, so I would really appreciate if someone reviewed it in case I missed something regardrless of the passed tests.
Thanks!
posix definitions should be used on Solaris
For compatibility reasons, Solaris historically had its header files
setup so that, unless specifically requested through specific header
defines, either the old pre-POSIX interfaces or POSIX.1c Draft 6
interfaces were used. However, in the case of rust, since these symbols
are linked directly instead of via system header files, the underlying
posix symbol name can be used directly instead.
These definitions should be corrected to match what they do on almost
every other platform.
Be aware this is a breaking change in terms of interface for any crates
/ consumers of these interfaces for Solaris.
Fixes#522
For compatibility reasons, Solaris historically had its header files
setup so that, unless specifically requested through specific header
defines, either the old pre-POSIX interfaces or POSIX.1c Draft 6
interfaces were used. However, in the case of rust, since these symbols
are linked directly instead of via system header files, the underlying
posix symbol name can be used directly instead.
These definitions should be corrected to match what they do on almost
every other platform.
Be aware this is a breaking change in terms of interface for any crates
/ consumers of these interfaces for Solaris.
Fixes#522
WIFCONTINUED function on Linux and FreeBSD
* Adds the missing WIFCONTINUED function to Linux (in `src/unix/notbsd/mod.rs`).
* Modifies the signature of WIFCONTINUED on FreeBSD (`src/unix/bsd/freebsdlike/mod.rs`) to be consistent with the ones on MacOS and Linux.
Correct solaris libc definitions:
* pthread_t is defined as uint_t, so must be c_uint, not uintptr_t, just
as pthread_key_t is already defined
* fd_set is defined as long, so must be i32/i64 based on
target_pointer_width; this also fixes an indirect endianness issue
encountered on sparc
* FD_SETSIZE should be defined as 65536 when target_pointer_width = 64
Fixes#515
* pthread_t is defined as uint_t, so must be c_uint, not uintptr_t, just
as pthread_key_t is already defined
* fd_set is defined as long, so must be i32/i64 based on
target_pointer_width; this also fixes an indirect endianness issue
encountered on sparc
* FD_SETSIZE should be defined as 65536 when target_pointer_width = 64
Fixes#515
Add AF_UNSPEC from Apple socket.h
[socket.h](https://opensource.apple.com/source/xnu/xnu-792.13.8/bsd/sys/socket.h) of Apple contains this definition so i'm not sure why it is missing from libc.
The same goes for other BSD-like OSes...
Is there some particular reasons why it is so?
