Initializing primitive from literal with cast

[gnzlbg]

I've seen code like this:

fn main() {
    let _ = 100 as usize;
    let _ = 100 as f64;
}

where clippy currently errors with

warning: casting i32 to f64 may become silently lossy if types change
 --> src/main.rs:3:13
  |
3 |     let _ = 100 as f64;
  |             ^^^^^^^^^^ help: try: `f64::from(100)`
  |
  = note: #[warn(clippy::cast_lossless)] on by default
  = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#cast_lossless

I think we should at least suggest if people didn't meant to write:

fn main() {
    let _ = 100_usize;
    let _ = 100_f64;
}

instead. Semantically, 100_i32 as f64 and 100_f64 are not the exact same thing, but in practice..

[rink1969]

I'd like to try this issue.
But I'm not sure what is expected output?
Is something like this OK?

warning: casting integer literal to f64 is unnecessary
 --> examples/clippytest.rs:3:13
  |
3 |     let y = 100 as f64;
  |             ^^^^^^^^^^ help: try: `100_f64`
  |
  = note: #[warn(clippy::unnecessary_cast)] on by default
  = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#unnecessary_cast

[gnzlbg]

cc @oli-obk I'm not sure whether suggesting 100_f64 only is ok, or whether both should be suggested.

[oli-obk]

Yes I think we should be extending the cast_lossless lint to suggest both solutions. This doesn't require a new lint.

[rink1969]

I think it's unnecessary cast for integer literal.
We can directly define f64 literal like 100_f64.

Another case is 100_i32 as f64, I also think it's unnecessary cast.
Same reason as before.
But this will break many test case.

[gnzlbg]

Is constructing a floating-point literal exactly the same thing as constructing and integer literal and casting it to a floating point value ? (e.g. are literals truncated according to the same rules as casts?).

[rink1969]

I checked the llvm ir generated by rustc.
100 as f64 100_f64 and 100_i32 as f64, these three cases got same irs( a floating-point literal).
f64::from(100) has an extra function call.

So I think the cast is unnecessary for integer literal.
But it's ok for var.

[rink1969]

I've submit a PR.
#3877
So we can see the effect about the fix.

[rink1969]

@gnzlbg Do you think we can close this issue?

[flip1995]

Yes this is implemented in #3877