Improve diagnostics for character literals containing multiple codepoints

[eggyal]

I tried to compile the following code (playground):

const SPADE: char = '♠️';

For the avoidance of doubt, ♠️ is U+2660. I expected the code to compile—but instead, the literal is rejected:

error: character literal may only contain one codepoint
 --> src/lib.rs:1:21
  |
1 | const SPADE: char = '♠️';
  |                     ^^^
  |
help: if you meant to write a `str` literal, use double quotes
  |
1 | const SPADE: char = "♠️";
  |                     ^^^

Notably, it is accepted if unicode-escaped:

const SPADE: char = '\u{2660}';

Meta

rustc --version --verbose:

rustc 1.57.0-nightly (e30b68353 2021-09-05)
binary: rustc
commit-hash: e30b68353fe22b00f40d021e7914eeb78473b3c1
commit-date: 2021-09-05
host: x86_64-apple-darwin
release: 1.57.0-nightly
LLVM version: 13.0.0

[SNCPlay42]

The spade "character" in your code is not just a single U+2660, it also has a unicode variation selector U+FE0F: playground

On my machine at least, this makes a visible difference in appearance (which github tries to eliminate by adding a U+FE0F unless I "escape" with a code block):

U+2660 U+FE0F: ♠️
U+2660 alone: ♠

[eggyal]

D'oh! Good spot, thanks @SNCPlay42. Perhaps some improved diagnostics could have helped here?

[SNCPlay42]

We have similar diagnostics in other cases of misleading Unicode so that seems like a good idea.