[breaking change] Cleanup linux and update MUSL

[gnzlbg]

• Update MUSL kernel headers to 4.4.2 (non-breaking)
• [breaking] MADV_SOFT_OFFLINE is not defined on MIPS
• [breaking] sendmmsg/recvmmsg take an unsigned int flag on MUSL
• [breaking] pthread_t is a pointer on MUSL
• rlimit resources should use a type alias on GNU (non-breaking)
• Deprecate SIGNUNUSED (should use SIGSYS instead)

[rust-highfive]

@gnzlbg: no appropriate reviewer found, use r? to override

[gnzlbg]

So IPV6_FLOWINFO is here: https://elixir.bootlin.com/linux/v5.0/source/include/uapi/linux/in6.h#L165
and that header "linux/in6.h" is included, but the tests fail with "undeclared ...". @mati865

[gnzlbg]

ah, duh, it appears that these are only included if the linux headers are included first -.-

[mati865]

I don't fully understand libc testsuite, it's a bit complicated for newcomer.
Anyway don't worry to ping me if you have issue with something.

[bors]

☔ The latest upstream changes (presumably #1346) made this pull request unmergeable. Please resolve the merge conflicts.

[gnzlbg]

I think the current way things are structured is quite weird, but maybe that's life.

For example, we have linux as a platform, and then musl, !musl+mips, !musl+s390x, and !musl (called other) as sub-modules. Where AFAICT !musl is just gnu. From the code !musl could also be uclib, newlib, etc. but since none of those are tested chances are none of those is working anyways.

So i'm tending towards just splitting the linux module into two modules: musl, and gnu, moving mips and s390x into the gnu submodule. If target_env is not musl or gnu for linux, we should error, and if there is someone using a different target_env on linux, they can just add it in parallel, instead of adding all sorts of workarounds to the gnu module.

[mati865]

Seems reasonable, chances for any other Linux libc to be fully compatible with gnu are close to zero.

[gnzlbg]

cc @asomers @alexcrichton - this enables most of the previously skipped tests on MUSL, fixing a couple of issues. Some of the breaking changes, like MADV_SOFT_OFFLINE not being available on MIPS, fix things that were already broken (this is not available on the latest 5.x kernel either). Other changes like pthread_t being a pointer on MUSL, could be breaking, but pthread_t should be handled like an opaque type by Rust code, so unless someone is manipulating these in ways they shouldn't, this should not break any code.

The only real breaking change is the argument type of sendmmsg/recvmmsg on MUSL, but AFAICT nix is not exposing these yet, and the std lib is not using them, so.. I hope we can get away with this one.

[gnzlbg]

@bors: r+

[bors]

📌 Commit 1a3d152 has been approved by gnzlbg

[bors]

⌛ Testing commit 1a3d152 with merge 48193e8...

[bors]

☀️ Test successful - checks-cirrus-freebsd-11, checks-cirrus-freebsd-12, checks-travis, status-appveyor
Approved by: gnzlbg
Pushing 48193e8 to master...

[apmanol]

Compilation for mips :
nix-0.14 for broke too.

info: component 'rust-std' for target 'mips-unknown-linux-gnu' is up to date
   Compiling nix v0.14.0
error[E0425]: cannot find value `MADV_SOFT_OFFLINE` in module `libc`
   --> /.cargo/registry/src/github.com-1ecc6299db9ec823/nix-0.14.0/src/sys/mman.rs:143:9
    |
143 |         MADV_SOFT_OFFLINE,
    |         ^^^^^^^^^^^^^^^^^ not found in `libc`
help: possible candidate is found in another module, you can import it into scope
    |
1   | use sys::mman::MmapAdvise::MADV_SOFT_OFFLINE;
    |

error[E0081]: discriminant value `14` already exists
   --> /.cargo/registry/src/github.com-1ecc6299db9ec823/nix-0.14.0/src/macros.rs:197:26
    |
197 |                   $entry = libc::$entry,
    |                            ^^^^
    |                            |
    |                            first use of `14`
    |                            enum already has `14`
    |
   ::: /.cargo/registry/src/github.com-1ecc6299db9ec823/nix-0.14.0/src/sys/mman.rs:105:1
    |
105 | / libc_enum!{
106 | |     /// Usage information for a range of memory to allow for performance optimizations by the kernel.
107 | |     ///
108 | |     /// Used by [`madvise`](./fn.madvise.html).
...   |
187 | |     }
188 | | }
    | |_- in this macro invocation

error: aborting due to 2 previous errors

[gnzlbg]

MADV_SOFT_OFFLINE not found

That's a nix error, that constant is not available on MIPS on any released Linux kernel.