Literal &[char] patterns clumsily require an as conversion

[jimblandy]

Many str methods accept patterns of any type that implements the std::str::pattern::Pattern trait. One such type is &[char], which matches any single character mentioned in the pattern slice.

However, &[char] patterns are not as nice to use as one might hope. This does not compile:

        let code = "\t    function noodle() { ";
        assert_eq!(code.trim_left_matches(&[' ', '\t']),
                   "function noodle() { ");

Rust complains:

error[E0277]: the trait bound `[char; 2]: std::ops::Fn<(char,)>` is not satisfied
   --> lib.rs:689:29
    |
689 |             assert_eq!(code.trim_left_matches(&[' ', '\t']),
    |                             ^^^^^^^^^^^^^^^^^ the trait `std::ops::Fn<(char,)>` is not implemented for `[char; 2]`
    |
    = note: required because of the requirements on the impl of `std::ops::FnOnce<(char,)>` for `&[char; 2]`
    = note: required because of the requirements on the impl of `std::str::pattern::Pattern<'_>` for `&[char; 2]`

The trim_left_matches method constrains its argument type P with P: Pattern, and Rust won't apply the usual type coercions like &[T; N] to &[T] in order to satisfy trait bounds.

One must instead write:

            let code = "\t    function noodle() { ";
            assert_eq!(code.trim_left_matches(&[' ', '\t'] as &[char]),
                       "function noodle() { ");

It's not a tragedy, but it's not pretty either.

CC @alexcrichton

[petrochenkov]

Shorter version: &[' ', '\t'][..]

[alexcrichton]

cc @Kimundi

[jimblandy]

Shorter version: &[' ', '\t'][..]

Lovely! This hadn't occurred to me.

[Kimundi]

Yeah, this was a known issue from the start.

This concrete impl is basically a originally-nice API that is old enough that it got gradually worse over time in multiple iterations. If I remember right, there was originally a hardcoded method that just took a &[char], and the language still allowed array literals to coerce to static slices, so originally you could do foo(['a', 'x', '0']). At some point we got a cleaning up of the expression syntax, which lead to foo(&['a', 'x', '0']). And then when I proposed the Pattern trait the methods got generic, and no longer liked to do deref coercions, leading to the ugly foo(&['a', 'x', '0'][..]) as shortest possible literal form.

I think the only real "fix" for this is deprecating this pattern, and/or holding out till we have char array generics and can have a Pattern impl for all [char; N] directly.

[steveklabnik]

Triage:

I think the only real "fix" for this is deprecating this pattern, and/or holding out till we have char array generics and can have a Pattern impl for all [char; N] directly.

Seems like the same status today; we are making progress on const generics, but it's not ready yet.