Non-ASCII identifiers should be restricted within extern blocks.

[crlf0710]

It was mentioned in rust-lang/reference#999 that in the original RFC text, there's text that said the
non-ASCII identifiers should be disallowed within extern blocks.

This is a niche use-case and i believe this doesn't need to block stablization. Creating this issue here to track it.

[jonas-schievink]

What will something like

extern {
    fn ö();
}

actually link against? This seems like it might be a forward-compatiblity hazard when linking to external code, since the precise mangling/encoding isn't clear

[Manishearth]

I think we should just disallow for now; and later check out what C does and match that.

[Manishearth]

@crlf0710 do you want to implement this? Should be a simple check in the parser.

[crlf0710]

@Manishearth Sure, let me give it a try.

[crlf0710]

Implemented in #83936 .

[tmiasko]

On the other hand the use of non-ASCII symbol names is already stable through link_name attribute.

[ogoffart]

Out of curiosity, I checked how Clang and GCC mangle the symbol: they simply use UTF-8.
MSVC doesn't seem to allow non-ascii in identifier.

extern "C" void héllo() {} // Mangle to "h\C3\A9llo"
void hélla() {} // mangle to "?h\C3\A9lla@@YAXXZ" or "_Z6h\C3\A9llav"

(Notice how the byte-length encoding in C++ is using the UTF-8 byte count, and not the code point count, as I would expect)
https://godbolt.org/z/bK8P71P46

[Manishearth]

Yeah I feel like the functionality can be re-added with an FCP without much work, there just needs to be some research on how stable those mangling schemes are around this.

[Manishearth]

As for link_name, my take on that is that once you're using that you're trying something weird and know what you're doing; there is no point restricting it further since it's very easy to accidently block out legit use cases.

[danielhenrymantilla]

Inversely, I've just noticed that despite usage of a properly-ASCII-ed link_name, if the "Rust name" is non-ASCII then it is still rejected:

extern {
    #[link_name = "cat"]
    fn ᓚᘏᗢ();
}

😿