Add warning control

[brson]

Do a few things:

• Turn on maximum warnings by default, in the compiler-specific way
• Add a method warnings_as_errors that turns warnings into errors in a compiler specific way

We haven't discussed this, but if we enable maximum warnings by default, it seems like some users will not want that and a way to turn it off. So it seems like we should also add a default_warnings method that declines to set the maximum warnings.

[opilar]

How to create a static array and return a slice of it?

   --> src/lib.rs:164:34
    |
164 |             ToolFamily::Msvc => &["/Wall"],
    |                                  ^^^^^^^^-
    |                                  |       |
    |                                  |       temporary value only lives until here
    |                                  does not live long enough
    |
    = note: borrowed value must be valid for the static lifetime...

Method:

    /// What the flags to enable all warnings
    fn warnings_flags(&self) -> &'static [&'static str] {
        match *self {
            ToolFamily::Msvc => &["/Wall"],
            ToolFamily::Gnu => &["-Wall", "-Wpedantic", "-Wextra"],
            ToolFamily::Clang => &["-Weverything"],
        }
    }

[alexcrichton]

Hm I thought it was like that but I guess not :(. Maybe try making this not a method but instead a temporary in a function?

[kornelski]

I'm unsure about it.

1. I mainly use this crate as implementation detail in *-sys crates, and users of my creates have absolutely no control over the code being compiled, so I don't want them to see any warnings they can't fix. Sometimes even I can't fix the warnings myself if I'm repackaging someone else's library. If warnings are on by default, it'll be a thing I always have to remember to turn off.

   I strongly prefer no warnings by default.

2. Combination of maximal warnings + warnings as errors is super brittle, and shouldn't be allowed. New lints introduced by newer compilers will break builds.

   It's dangerous in general, because different compilers may warn about different things, so authors testing on clang may break crates for MSVC users just because MSVC complains about different things. I remember struggling with basic things like unused variables, where (void)var; works to silence unused warning in one compiler, but in other compiler causes useless expression warning.

[dtolnay]

I agree with @pornel. Warnings enabled seems like the wrong default for the way this crate is usually used, and I would never want to use warnings-as-errors.