new lint: unused_unit

[llogiq]

This is the rejected unused_parens extension plus a lint for units in function return types and return expressions.

It appears to fail to obey the allow(..) attributes though (see copies.rs) – does anyone know why that happens?

[flip1995]

Maybe the ignored allow(..) is related to #3198. Something weird is going on with defining lint levels. I didn't find the time to investigate this further yet.

[llogiq]

Maybe the lint level selection was broken when tools lints were added?

[flip1995]

I thought the same, so I searched for merge commits by @bors that broke the lint level definition and I got it down to 3 commits: #3198 (comment). None of these commits are related to tool_lints though..

[llogiq]

I have extended the lint to cover explicit unit return types in fn types and Fn* trait bounds. Apart from the lint level selection issue, this should be ready to merge.

[llogiq]

I had missed adding the lint to the lint-array, which resulted in the failing allow annotation. This should hopefully now work. r? once CI is happy.

[llogiq]

@oli-obk r?

[mati865]

Maybe too pedantic but there is no dot at the end of the line.

[aldanor]

@llogiq This seems to act weird in cases like this:

thread_local!(static A: () = ());

yields

warning: unneeded unit expression
   |
20 | thread_local!(static A: () = ());
   |                              ^^ help: remove the final `()`

... and allow(clippy::unused_unit) doesn't seem to work in this case (only if done at the module level).

[llogiq]

True, that looks weird. Then again, why do you need a zero-sized thread-local?

[aldanor]

Then again, why do you need a zero-sized thread-local?

You'd be surprised :) For identifying threads when implementing a reentrant mutex.