Improve dead code analysis for structs and traits defined locally

[mu001999]

This PR does some refactor and improvement on the dead code analysis, and doesn't lint pub structs.

1. refactors the two-phase check for impls and impl items of trait
   1. checks them all later because we must use the trait/trait item and the adt defined locally firstly
   2. makes the logic cleaner and doesn't require special checks about whether it's public or not
2. mark the adt live if it appears in pattern, like generic argument, it also implies the use of the adt
   1. based on 1 and 2, we can detect unused private adts impl Default, without adding special logics for Default
   2. so that we can remove rustc_trivial_field_reads on Default, and the logic in should_ignore_item
3. extends rules to impls for types which refer to adts, like &Foo/[Foo] things
4. lints unused assoc consts like assoc fns, and unused traits with assoc tys by marking assoc tys live only if the trait is live (is same to Mark assoc tys live only if the corresponding trait is live #126618 reverted in Revert recent changes to dead code analysis #128404)

Fixes #120770
Fixes #126729
Fixes #127911
Fixes #128839

cc @pnkfelix
r? @compiler-errors

[rustbot]

Some changes occurred in tests/ui/sanitizer

cc @rust-lang/project-exploit-mitigations, @rcvalle

[compiler-errors]

Please separate this into separate commits each implementing an individual tweak to the dead code analysis, with the tests adjusted at each commit so I can see the fallout from each change specifically. It's very difficult to map the changes to the UI tests to each code change without that. --- I want to think very hard about each of these changes to determine if there are any false positives that are caused by each change, and that is harder to do with a single commit.

Also, if you want, please open a separate PR that removes the dead code from the compiler/standard library that is now detected after these changes. That can be landed separately.

[mu001999]

@compiler-errors I have separated this to separate commits.

[bors]

☔ The latest upstream changes (presumably #128761) made this pull request unmergeable. Please resolve the merge conflicts.

[bors]

☔ The latest upstream changes (presumably #129046) made this pull request unmergeable. Please resolve the merge conflicts.

[cjgillot]

Could you submit 1317d54 as a separate PR?

[mu001999]

It is probably necessary to make the lint handle associated types, I'd like an explanation why it is necessary

Just detecting more dead codes? I think removing more unused items is helpful. And we have detected unused inherit assoc types, although the feature is unstable for now.

and why we can't just rely on HIR and typeck-output resolutions.

We can only get Res::Err when meeting a assoc type in hir. typeck_results.qpath_res works but it only work for QPaths in an Expr or Pat node. And we may have cases like:

trait TwoWayStrategy {
    type Output;
    fn use_early_reject() -> bool;
    fn rejecting(a: usize, b: usize) -> Self::Output;
    fn matching(a: usize, b: usize) -> Self::Output;
}

I asked such things before in zulip

[mu001999]

@rustbot ready

[bors]

☔ The latest upstream changes (presumably #129595) made this pull request unmergeable. Please resolve the merge conflicts.

[bors]

☔ The latest upstream changes (presumably #129941) made this pull request unmergeable. Please resolve the merge conflicts.

[mu001999]

@cjgillot friendly ping

[mu001999]

It is probably necessary to make the lint handle associated types, I'd like an explanation why it is necessary

@cjgillot

Updated: using simpler rules for assoc tys instead of reasoning on types, i.e., only marking assoc tys live only if the corresponding trait is live, which can help to lint more unused traits (is same to #126618 reverted in #128404).

Removes old logic for unused assoc tys.
I think the it needs further consideration and could later be made as a separate PR.

[wesleywiser]

Hi @cjgillot, it looks like this PR is ready for review again when you have time. Thanks 🙂