`#![allow(clippy::` incorrectly completes with a leading `clippy::`

[RDambrosio016]

Reproducible code

#![allow(clippy::

Select any of the lints, it will be completed as #![allow(clippy::clippy::or_fun_call)] for example, which is incorrect.

I presume this is because vsc does not consider clippy:: to be a full ident per the rust grammar, therefore it thinks the completion should go after it because the first part isnt part of it

[crumblingstatue]

I noticed that what the user typed so far is highlighted in the completion hints

However, if :: is typed after clippy, the highlighting stops working.

Is the above assessment correct in this being a fault of visual studio code? Should it be fixed there, or can it be fixed in rust-analyzer?

EDIT: Actually, the highlighting behavior is the same for other kinds of completions after double colon. What's different is that the completion hints don't include the part before the ::

I did some debug printing inside complete_attribute(acc: &mut Completions, ctx: &CompletionContext), and I noticed that when doing autocomplete after ::, for regular paths, ctx.original_token is COLON2@285..287 "::", but for lint attributes, it's COLON@32..33 ":". Looks like for some reason it's not being recognized as COLON2 in lint attributes.

[bjorn3]

Something similar happens with pub(crate) which turns into pub(pub(crate)).

[crumblingstatue]

Looks like the attribute is being parsed as

ATTR@0..18
      POUND@0..1 "#"
      L_BRACK@1..2 "["
      META@2..17
        PATH@2..7
          PATH_SEGMENT@2..7
            NAME_REF@2..7
              IDENT@2..7 "allow"
        TOKEN_TREE@7..17
          L_PAREN@7..8 "("
          IDENT@8..14 "clippy"
          COLON@14..15 ":"
          COLON@15..16 ":"
          R_PAREN@16..17 ")"
      R_BRACK@17..18 "]"

Shouldn't that bet a COLON2 instead of 2 COLONs?

[RDambrosio016]

Macro invocations are raw token trees, so no tokens are joined because there is no "semantic meaning"

[crumblingstatue]

Welp, I give up for now, haha. I have no idea how to debug this.

[RDambrosio016]

I think this is just an issue with vsc not thinking the colons form a full ident and therefore putting the completion just after the colons, while RA should be telling it to replace everything in the completion

[Veykril]

The problem here is that we do not recognize the clippy:: prefix in the completion building process. So when we insert the completion we insert if after the already existing clippy:: instead of replacing it.

Basically the ctx.source_range() call here is incorrect, instead we need to fetch the ranges of the clippy and :: token as ctx.source_range() returns the range of the current identifier we are completing, which isn't correct in this case. https://github.com/rust-analyzer/rust-analyzer/blob/9f24cc7126f6508dcc9e429a33bd12975873d3e4/crates/ide_completion/src/completions/attribute/lint.rs#L24

[Veykril]

Fixing this is rather tricky as we have to basically manually reparse the input we are expecting. It would be great if we could reuse that path completion infra here but inputs to attributes are just random tokens to us.