Refresh the CI setup

[Lokathor]

This is its own PR just to fix the CI/features, we can do the other thing with sqrt separately once we're confident we have this part fully sorted out.

• Remove the checked feature, just use the debug_assertions flag that's already part of Rust
• Swap the stable feature when we want Stable to be an unstable feature when we want Nightly bonuses.
• There are no longer any default features. The crate should operate in a totally plain mode by default.
• Since #![deny(warnings)] has been in the lib.rs the whole time, and since my editor runs clippy on the source, I at least silenced all the clippy warnings that we definitely don't care about. There's a lot more to do on the clippy front, but that means doing little tiny changes to almost every single file, so we'll leave that out for now

---

I'm putting up this PR so people can look and I can start to see where the CI errors are, I really don't expect it to pass CI this first time through.

[alexcrichton]

Looks reasonable to me!

[Lokathor]

• The PowerPC build seems to have failed because of a crates.io fluke, because 0.6.5 is a valid version of rand and all, https://crates.io/crates/rand/0.6.5
• The ARM build seems to be something to do with the jn function, but as far as I can tell that file wasn't affected by any change in the PR so far so I don't know why it hit a failure.

Can anyone else tell what's wrong with jn?

[gnzlbg]

Can anyone else tell what's wrong with jn?

Apparently it gives the wrong answer:

'INPUT: [140, 4603707074953425158] EXPECTED: [24658263201] ACTUAL 0.0'

You might want to open a bug. I'll retry this, that input is generated randomly.

[alexcrichton]

There's a known issue with jn and jnf which would explain that.

[Lokathor]

Alright, then I'm marking this as ready for review and merge

[alexcrichton]

Oh right one last thing before merging, @Lokathor would you be up for updating the submodule in rust-lang-nursery/compiler-builtins and getting CI passing there as well?

[Lokathor]

Uh, I guess? I'm not familiar with how to make it run CI against an un-released version.

Which reminds me that this needs to be a breaking version bump for this libe since it's a big breaking change.

[Lokathor]

oh silly me, 0.2.0 isn't released on crates.io yet so this can all still be 0.2.0

[Lokathor]

Oh it's a git submodule. I thought it was like a Cargo thing.

I'm afraid I can't help with that. Every time I do anything more than just commit and push things go badly for me. I don't know how git submodules work at all.

[alexcrichton]

Yes I basically mean just send a PR to the compiler-builtins repo and get CI green there. You can update the submodule by editing .gitmodules to your git repository and then going into the folder and resetting the git repo to the revision of this PR.

[Lokathor]

i can at least give it a try later today

[Lokathor]

... so I tried to use git reset on a clone of the repository but it didn't appear to do anything

D:\dev\compiler-builtins>git reset

D:\dev\compiler-builtins>cd libm

D:\dev\compiler-builtins\libm>git reset

D:\dev\compiler-builtins\libm>

(Unfortunately I genuinely don't know how git works at all, I just use the Windows GUI thing and that's the limit of my git skills.)

I think whatever git magic you want tested has to be done by someone else.

[alexcrichton]

A git submodule is a url and a sha for a commit, so you can update the URL by editing .gitmodules and then you need to check the actual submodule to a particular commit to commit it within the parent repository. That means you'll need to go into the libm submodule, add a remote necessary to fetch your branch, reset it to this commit, and then commit everything. That probably looks something like:

$ cd compiler-builtins
# edit .gitmodules
$ cd libm
$ git remote add lokathor https://github.com/Lokathor/libm
$ git fetch lokathor
$ git reset --hard lokathor/new-CI
$ cd ..
$ git add libm .gitmodules
$ git commit -m "Update the libm submodule"

[Lokathor]

The current file has no commit hash in it, https://github.com/rust-lang-nursery/compiler-builtins/blob/master/.gitmodules,

But I did the other steps and opened the PR anyway, rust-lang/compiler-builtins#313

[alexcrichton]

Ok looks like the compiler-builtins PR is green, so let's merge!