document behavior of \r\n in string literals

[matklad]

No description provided.

[matklad]

Note that the current implementation works by converting \r\n -> \n at the file loading time, but I think the current wording is equivalent to the implementation.

[Havvy]

💟 Thanks!

[ehuss]

This looks good, but I was wondering (as noted in #626) if we should instead note how CRLF translation happens in one place. As it stands, it's not clear how other things like byte literals handle it. If it is noted in one place, then it wouldn't need to be repeated anywhere else. Does that sound like a good idea?

I haven't delved too deep into it yet, but the things I'd like to see:

• does all source text unconditionally translate CRLF to LF?
• how exactly bare CRs are handled.
• what is the interaction with include_str!/include_bytes!/include! (presumably that would be in the std docs).

[matklad]

@ehuss yeah, that makes sense. I'm not sure I'll manage to send a follow-up PR, so here's an unedited summary of things:

does all source text unconditionally translate CRLF to LF?

Yes, with only two possible exceptions of include_bytes and include_str.

This in general is not observable, except for literals and doc comments.

how exactly bare CRs are handled.

They are not translated, \r\r\n turns into \r\n, and the lexer treats only \n as a new line. However, we treat isolated \r in a string literal as an error. Argubly, the latter should be downgraded to a warning, but also warn about other control characters: rust-lang/rust#63682