switch to a current nightly

[japaric]

No description provided.

[japaric]

@homunkulus r+

[homunkulus]

📌 Commit 4df5deb has been approved by japaric

[homunkulus]

⌛ Testing commit 4df5deb with merge 4df5deb...

[homunkulus]

💔 Test failed - status-travis

[japaric]

@homunkulus try

[homunkulus]

⌛ Trying commit 5908bb7 with merge 5908bb7...

[homunkulus]

☀️ Test successful - status-appveyor, status-travis
State: approved= try=True

[japaric]

@mattico so, this is interesting. We are currently testing the arm-unknown-linux-gnueabi target with the nightly from 2016-09-21 and the all the tests pass. But in this PR, I tested the same arm target but vs today's nightly and it produces the failures you are seeing in #90 (see above, the first travis failure).

Now, the .travis.yml has a comment that says that we are using the older nightly for arm because of rust-lang/rust#36518 which, iirc, is the problem with u32 -> f32 u32 -> f64 being UB because those are intrinsics implemented in compiler-rt. Now, that particular issue supposedly has already been already fixed (rust-lang/rust#36828) but perhaps the fix is not yet in today's nightly.

[japaric]

@mattico Some things you could try to fix #90. Avoid doing a integer -> float (e.g. u32 -> f64) conversion. Maybe try directly generating random floating point values. Otherwise, you could try porting floatsidf and floatdidf to Rust so these tests end up using the Rust implementation rathern than compiler-rt's.