Get nix passing mips tests

[berkowski]

Addresses #533

Working on getting nix to pass mips-unknown-linux-gnu and mipsel-unknown-linux-gnu

Includes a few commits to disable non-related failing tests. These will be reverted before merge.

Still fails two tests related to mq_setattr:

---- test_mq::test_mq_set_nonblocking stdout ----
	thread 'test_mq::test_mq_set_nonblocking' panicked at 'called `Result::unwrap()` on an `Err` value: Sys(EPERM)', /buildslave/rust-buildbot/slave/stable-dist-rustc-cross-rustbuild-linux/build/src/libcore/result.rs:837

---- test_mq::test_mq_setattr stdout ----
	thread 'test_mq::test_mq_setattr' panicked at 'called `Result::unwrap()` on an `Err` value: Sys(EPERM)', /buildslave/rust-buildbot/slave/stable-dist-rustc-cross-rustbuild-linux/build/src/libcore/result.rs:837

These will PASS if you override the value of libc::O_NONBLOCK to use the x86_64 value of 2048:

diff --git a/src/mqueue.rs b/src/mqueue.rs
index 2c12750..4024698 100644
--- a/src/mqueue.rs
+++ b/src/mqueue.rs
@@ -133,7 +133,7 @@ pub fn mq_setattr(mqd: mqd_t, newattr: &MqAttr) -> Result<MqAttr> {
 /// Returns the old attributes
 pub fn mq_set_nonblock(mqd: mqd_t) -> Result<(MqAttr)> {
     let oldattr = try!(mq_getattr(mqd));
-    let newattr = MqAttr::new(O_NONBLOCK.bits() as c_long,
+    let newattr = MqAttr::new(2048 as c_long,
                               oldattr.mq_attr.mq_maxmsg,
                               oldattr.mq_attr.mq_msgsize,
                               oldattr.mq_attr.mq_curmsgs);
diff --git a/test/test_mq.rs b/test/test_mq.rs
index fd050d4..b911e1e 100644
--- a/test/test_mq.rs
+++ b/test/test_mq.rs
@@ -83,7 +83,7 @@ fn test_mq_setattr() {
     // O_NONBLOCK can be set (see tests below)
     assert!(new_attr_get.unwrap() != new_attr);
 
-    let new_attr_non_blocking =  MqAttr::new(O_NONBLOCK.bits() as c_long, 10, MSG_SIZE, 0);
+    let new_attr_non_blocking =  MqAttr::new(2048 as c_long, 10, MSG_SIZE, 0);
     mq_setattr(mqd, &new_attr_non_blocking).unwrap();
     let new_attr_get = mq_getattr(mqd);
 
@@ -101,7 +101,7 @@ fn test_mq_set_nonblocking() {
     let mqd = mq_open(mq_name, O_CREAT | O_WRONLY, S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH, Some(&initial_attr)).unwrap();
     mq_set_nonblock(mqd).unwrap();
     let new_attr = mq_getattr(mqd);
-    assert!(new_attr.unwrap().flags() == O_NONBLOCK.bits() as c_long);
+    assert!(new_attr.unwrap().flags() == 2048 as c_long);
     mq_remove_nonblock(mqd).unwrap();
     let new_attr = mq_getattr(mqd);
     assert!(new_attr.unwrap().flags() == 0);`

I don't understand this at the moment. Perhaps it's a bug with trust/cross? The value in libc matches current linux headers and their own checks.

[posborne]

@alexcrichton @japaric Any ideas on this one?

[alexcrichton]

Unfortunately no, I've never used these APIs :(

[berkowski]

I'm able to replicate these errors and behaviour using C and qemu-mips 2.6.1 and so I don't think there's an issue specific to trust or cross. Whatever is going on here seems outside of rust's scope and getting pretty far beyond my own knowledge.

I can clean up this pull as is -- it does fix a lot of other compile issues.

[posborne]

I can clean up this pull as is -- it does fix a lot of other compile issues.

Yes, let's move forward with that. The issues may be specific to the kernel/libc being used in the MIPS QEMU VM.

[berkowski]

I removed the two commits modifying other test and travis behaviour. If you want the whole thing squashed let me know.

[posborne]

If you want the whole thing squashed let me know.

This looks fine to me; no need.