Add a lint for unused type parameters

[sanxiyn]

Fix #25871.

[rust-highfive]

r? @nikomatsakis

(rust_highfive has picked a reviewer for you, use r? to override)

[huonw]

It seems weird that this is necessary? Where are things unused?

[sanxiyn]

One example is core::intrinsics::size_of<T>.

[huonw]

Oh, hm. Shouldn't extern fns with type params consider them used? (But, in general, it makes sense that there may be a few things like that.)

[alexcrichton]

It looks like there's a number of legitimate use cases for having "unused" parameters, so can this be an allow-by-default lint instead of warn-by-default?

[sanxiyn]

The reason this is a lint (and not a hard error, unlike struct S<T>) is because there are legitimate use cases. But they are rare, and in my opinion rare enough that writing #[allow(unused_type_parameters)] is acceptable.

[alexcrichton]

I understand my viewpoint on whether lints should be warn by default is subject, by my believe is that any false positive on a lint means that it should not be warn-by-default. I've found that over time projects end up gaining dozens of little #[allow] directives that end up turning into spaghetti because each lint we have has very few false positives but the union of all the false positives means that you're going to have a lot.

[sanxiyn]

This is not the first time you argued for allow-by-default. In #10477 you wrote:

There is a much larger use of #[allow(dead_code)] than I would expect throughout the codebase. I think this may be a sign that this should be an allow-by-default lint.

But dead_code lint was accepted warn-by-default. I am curious what changed your mind, because like @huonw on that thread, I think this should be warn-by-default.

[alexcrichton]

This is not the first time you argued for allow-by-default.

I agree! I have historically been very anti-lints as I feel we have far too many already and generate too many warnings for false positives.

But dead_code lint was accepted warn-by-default. I am curious what changed your mind, because like @huonw on that thread, I think this should be warn-by-default.

Right, but that was almost two years ago. I would want it to be allow-by-default today.

[nikomatsakis]

This is an interesting discussion. I wonder if we should adopt the convention that you can write _T: Foo to silence the lint, analogous with unused variables?

[bors]

☔️ The latest upstream changes (presumably #27818) made this pull request unmergeable. Please resolve the merge conflicts.

[alexcrichton]

Closing due to inactivity, but feel free to resubmit with a rebase!