Tuple indexing should handle float tokens specially

[chris-morgan]

tuple.0.0 tokenises as [Ident("tuple"), Dot, Float("0.0")]. At present it produces an “unexpected token” error with a help message along the lines of “try parenthesizing the first index; e.g., (tuple.0).0”.

This is silly. We should just special-case float literals of the form /\d+.\d+/ and understand them sensibly—that the original one be treated as [Ident("tuple"), Dot, Int("0"), Dot, Int("0")]. It’s not as though there’s any ambiguity (that I see, anyway), and it’s much nicer to handle something automatically where there is no ambiguity and the meaning is perfectly clear, rather than producing a helpful error message.

There’s precedent for this approach to things in the handling of the && token, where the expression &&foo and the pattern &&foo are both treated as two & tokens rather than as a single && token.

[huonw]

NB. this behaviour was explicitly discussed as part of rust-lang/rfcs#184 that designed this feature.

[steveklabnik]

/cc @rust-lang/lang, but yes, I thought this was explicitly decided against in the RFC

[chris-morgan]

(I guess #20287 and #23694 are the same as this; I just didn’t find either of them in my searching.)

I really think we should reconsider this. I don’t get the feel that it was so much explicitly decided against in the RFC as that one or possibly two people felt uncomfortable with it and so it was removed from the RFC as the easiest thing to do.

I really don’t see any reason for not doing this; it’s not difficult to do, and it makes things do what people expect, as is evidenced by the very way in which the special help message on the error is done. Can anyone present an argument for not doing it? @nikomatsakis, as you preferred not to change the lexer thus at that time a year ago?

[liigo]

+1

[nikomatsakis]

On Sat, Sep 05, 2015 at 02:01:02PM -0700, Steve Klabnik wrote:

/cc @rust-lang/lang, but yes, I thought this was explicitly decided against in the RFC

It was, though I've been bit by it a few times and wondered if we made
the right call from time to time.

[pnkfelix]

(and @nikomatsakis did even say at that time "we can always add it later" :) )

[hayathms]

I think Adding it is better
plain chained dot notation for accessing nested tuple is required as it feels more natural.

tuple_of_tuples.1.1.0.3.0.1 -> heart says it should work :) :) :)

[steveklabnik]

Triage: this was an explicitly made decision in an RFC, and so I'd argue that it needs an RFC to change. Given that and the lack of comments, I'm going to give this one a close.

[dtolnay]

I put up a fix in #70420. I don't think an RFC is going to be required. If the compiler team is on board with the approach, then the lang team can do FCP on the PR.