dropck doc oversights / revision

[pnkfelix]

The implementation of non-parametric dropck (#28861) included comments and doc that is missing a few things.

1. There was feedback from @gankro on changes to the rustonomicon that should be further tweaks (see comment thread from implement RFC 1238: nonparametric dropck. #28861)
2. There should be a huge warning (maybe even a lint or something) that if you put #[unsafe_destructor_blind_to_params] on a struct that is using native pointers (e.g. struct Foo<T>( { ptr: *mut T }) then you almost certainly need a PhantomData in there.

The old code before #28861 was in fact busted in this respect, as one can see by taking the test case from #29106 and running it on older versions of Rust.

[Gankra]

👍 for a lint; CC @Manishearth

[Manishearth]

hardwired lint makes sense here.

Note that if you add a lint, you need to be careful about backcompat if you're going to be removing it. We don't have a framework for "obsolete" lints yet, but it could be added.

[steveklabnik]

Triage: I didn't even know this attribute was a thing!

[apasel422]

@steveklabnik The unsafe_destructor_blind_to_params attribute is deprecated now. CC #34761.

[Mark-Simulacrum]

@pnkfelix Is there a chance you could comment on the status of this issue since the attribute was deprecated in favor of (other?) issues.

[steveklabnik]

Triage: with no comment from @pnkfelix, I'm going to consider this stale and close it. felix, please let me know if you think this is still important and we can re-open.

[pnkfelix]

I imagine the same scenario arises with #[may_dangle], which is what replaced unsafe destructor blind to params