Implement Debug for EncodeWide

[thaliaarchi]

Since std::os::windows::ffi::EncodeWide is reexported from std::sys_common::wtf8::EncodeWide, which has #![allow(missing_debug_implementations)] in the parent module, it does not implement Debug.

Implement it like core::str::Chars. Format the WTF-16 code units as char literals and escape them when they're surrogate halves. Although this is not valid Rust syntax, this is fine for Debug.

This becomes insta-stable.

r? libs-api

[bjorn3]

We can't format it like char, because \u escape sequences for surrogate halves are invalid syntax in Rust.

Even if rust doesn't accept \u for surrogate halves, we can still format them as \u. Debug isn't guaranteed to be valid rust source code anyway.

[thaliaarchi]

Here are the options, as I see them. I think formatting them as chars when not surrogates is the most readable for most texts. Rust style seems to lean towards uppercase hex.

1. EncodeWide([97, 233, 32, 55357, 55357, 56489]): decimal
2. EncodeWide([0x61, 0xE9, 0x20, 0xD83D, 0xD83D, 0xDCA9]): hex, zero-pad min 2, upper
3. EncodeWide([0x61, 0xe9, 0x20, 0xd83d, 0xd83d, 0xdca9]): hex, zero-pad min 2, lower
4. EncodeWide([0x0061, 0x00E9, 0x0020, 0xD83D, 0xD83D, 0xDCA9]): hex, zero-pad 4, upper
5. EncodeWide([0x0061, 0x00e9, 0x0020, 0xd83d, 0xd83d, 0xdca9]): hex, zero-pad 4, lower
6. EncodeWide(['a', 'é', ' ', '\u{D83D}', '\u{D83D}', '\u{DCA9}']): pseudo-char, escaped surrogates, upper
7. EncodeWide(['a', 'é', ' ', '\u{d83d}', '\u{d83d}', '\u{dca9}']): pseudo-char, escaped surrogates, lower

[thaliaarchi]

I've updated formatting to be 4-wide zero-padded hex (style 4)

For another opinion, @ChrisDenton wrote:

I think EncodeWide, semantically, produces a string like a str but with a different encoding. That said, EncodeWide itself is an iterator more akin to a Bytes iterator of &str (except u16 instead of bytes). But this is all made more complicated by the fact that std doesn't have a WideStr type so people typically collect into a Vec<u16>, which doesn't know it's a wide string. Whereas the iterator does know. So I am sympathetic to the char like representation but for debug I do think we really have to go with a hex repr. And my gut preference is for the 4 zero pad, which is more suggestive of unicode without introducing illegal rust syntax. But I reserve the right to change my mind on that 😆

I think the char-like repr is a non-starter in any case because it looks like rust code but may be illegal (i.e. for surrogates).

(in favor of style 4)

[thaliaarchi]

What's needed next to move this along?

[ChrisDenton]

I've nominated for libs-api. I don't know if it needs a full libs API discussion but I believe at least someone from the team will need to sign off on it.