The useless attribute lint doesn't put the ! in the right place.

[BookOwl]

When the useless_attribute lint is activated, it says that you most likely forgot an exclamation mark (!). While this is true, it also suggests to stick the ! in the middle of the attribute:

warning: useless lint attribute, #[warn(useless_attribute)] on by default
 --> src/lib.rs:1:38
  |
1 | #[cfg_attr(feature = "cargo-clippy", allow(new_without_default))]
  |                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^
  |
help: if you just forgot a `!`, use
  | #[cfg_attr(feature = "cargo-clippy", a!llow(new_without_default))]
  = help: for further information visit https://github.com/Manishearth/rust-clippy/wiki#useless_attribute

The correct response should be to suggest adding it after the #
I'm using the latest clippy on the latest nightly Rust.

[TimDiekmann]

I just had this issue with enum_glob_use too. I guess, the message is bugged in general.

[phansch]

It looks like there is no attribute information for cfg_attr (item.attrs does not include it at all). I will take a stab at this and see if I can find a workaround.

I believe rust-lang/rust#25544 is related somehow

[phansch]

Unfortunately this seems really difficult to find a proper fix for. Since cfg_attr and cfg attributes are not accessible in the linting process, the only way to work around that, would be to use the CodeMap to search for the beginning of the attribute.

I managed to get a fix working, but only if the whole cfg_attr attribute is on a single line. Cases where the cfg_attr attribute is split over multiple lines would still be broken. I'm not sure if it makes sense to parse the whole file ourselves and look for the beginning of each attribute that spans multiple lines.

Maybe it's easier to remove the suggestion from this lint for now? @oli-obk

[oli-obk]

@phansch can you detect the cfg case? if so, just don't emit the suggestion in the cfg case

[phansch]

Sorry, my previous comment was partly wrong (now updated):

As far as I can tell, only cfg_attr is missing from the item attributes, so I think we don't have to worry about cfg. Using CodeMap, I can detect cfg_attr somewhat easily if it's on the same line, but it's more work to detect if the attribute starts on a previous line, like

#[cfg_attr(feature = "cargo-clippy",
  allow(dead_code),
  allow(new_without_default))]

That would require iterating over each line towards the beginning of the file, making a Span out of each Line, checking if the span contains the beginning of the attribute (a #[), if yes, replacing it with #![.

cfg is included in the item attributes, but I'm not sure if it matters for this lint, because as it is right now, cfg attributes are not linted in this lint.

[oli-obk]

So... we could just not create a suggestion at all in the multi line case?

[phansch]

That would be the easier solution, yes. If the attribute starts on the same line, we show the correct suggestion, otherwise we don't show a suggestion 👍