Rustup to rust-lang/rust#64874

[flip1995]

TODO:

• replace rvalue_promotable_map in [1]
• fix [2] according to this comment Simplify ExprUseVisitor rust#64874 (comment) this should be merged with consume, but I didn't figure out how to merge them, yet.
• fix [3]; What to do with LoanCause?

[2]+[3] probably have to be resolved by a rewrite of the lint. #4628 (comment)

[1]

rust-clippy/clippy_lints/src/methods/mod.rs

Lines 1292 to 1299 in 54bf4ff

	let promotable = self
	.cx
	.tcx
	.rvalue_promotable_map(owner_def)
	.contains(&expr.hir_id.local_id);
	if !promotable {
	self.found |= true;
	}

[2]

rust-clippy/clippy_lints/src/escape.rs

Line 126 in 54bf4ff

fn consume_pat(&mut self, consume_pat: &Pat, cmt: &cmt_<'tcx>, _: ConsumeMode) {

[3]

rust-clippy/clippy_lints/src/escape.rs

Lines 166 to 176 in 54bf4ff

	fn borrow(
	&mut self,
	_: HirId,
	_: Span,
	cmt: &cmt_<'tcx>,
	_: ty::Region<'_>,
	_: ty::BorrowKind,
	loan_cause: LoanCause,
	) {
	if let Categorization::Local(lid) = cmt.cat {
	match loan_cause {

I could need some help with [1]. The purpose of this is to "don't lint for constant values". cc @matthewjasper

For now I see what I can do with [2].

changelog: Temporary break boxed_local lint.

[flip1995]

Is removing this completely ok? It didn't break anything in the tests 🤔 cc @sinkuu since you added this function in 627d24c

[sinkuu]

Nothing is inserted to spans_need_deref after this change, so the suggestions to add * should be broken, but idk.

Anyway, Rust has gained default binding mode and this suggestion became useless. I think it's OK to remove it.

[flip1995]

Update: I won't get to this until tomorrow. It would be great if someone could take over :)

[matthewjasper]

[1] Change is_ctor_function to check for promotable functions as well (and give it a different name). It would look something like

// ...
match cx.tables.qpath_res(qp, fun.hir_id) {
    def::Res::Def(DefKind::Variant, ..) | Res::Def(DefKind::Ctor(..), _) => true,
    // Maybe consider `is_const_fn`, but there's less guarantee of cheapness there.
    def::Ref::Def(DefKind::Fn, def_id) => cx.tcx.is_promotable_const_fn(def_id),
    _ => false,
}

[2] and [3] Looking into this lint some more, it's hopelessly broken and needs a rewrite. That said, to roughly keep the current behavior:

[2]:

• Move the if is_argument part into consume, using cmt.hir_id instead of consume_pat.hir_id.
• Remove the if let Categorization::Rvalue(..) = ... block, it doesn't appear to work.
• Move the if let Categorization::Local(lid) = ... block into consume adding a check that checking that map.find(cmt.hir_id) is Some(Node::Binding(_))

[3]: Unconditionally call self.set.remove(&lid); - this slightly changes the behavior of this lint around closures.

[flip1995]

Thanks a lot for the hints @matthewjasper!

Oh nevermind...

u32::max_value() (and so on) is a DefKind::Method, not DefKind::Fn

[1] That doesn't work, because the PathSegment of a const fn call like u32::max_value() always has res: Some(Err), e.g.

let _ = 1u32.checked_add(1).unwrap_or(u32::max_value());

dbg!(qpath):

[clippy_lints/src/utils/mod.rs:809] qp = TypeRelative(
    type(u32),
    PathSegment {
        ident: max_value#0,
        hir_id: Some(
            HirId {
                owner: DefIndex(17),
                local_id: 43,
            },
        ),
        res: Some(
            Err,
        ),
        args: None,
        infer_args: true,
    },
)

I'm still curious about this though:

What does Res::Err mean? That the resolution errored or that no resolution is known at this point? (Does Res even stand for Resolution?) Sorry if that's a stupid question, but def::Res is completely undocumented and I couldn't find anything in the rustc-guide.

---

it's hopelessly broken and needs a rewrite

That's what I was afraid of. I propably won't have the time of doing a rewrite, so I'd suggest to just adapt the tests, since this only introduces FNs. We then can get a working Clippy faster and work on this later. cc @rust-lang/clippy

[2]+[3] I did this is in b7d4735, but the result is, that all tests fail for this lint

[flip1995]

On another note: I think it's time to Nuke the clippy plugin interface

warning: use of deprecated attribute `plugin_registrar`: compiler plugins are deprecated. See https://github.com/rust-lang/rust/issues/29597
  --> src/lib.rs:12:1
   |
12 | #[plugin_registrar]
   | ^^^^^^^^^^^^^^^^^^^ help: remove this attribute
   |
   = note: `#[warn(deprecated)]` on by default

https://travis-ci.com/rust-lang/rust-clippy/jobs/242398220#L890-L896

[flip1995]

As suggested in #4616 (comment)

@bors treeclosed=10 p=10

[matthewjasper]

What does Res::Err mean? That the resolution errored or that no resolution is known at this point? (Does Res even stand for Resolution?) Sorry if that's a stupid question, but def::Res is completely undocumented and I couldn't find anything in the rustc-guide.

Res::Err usually means that name resolution errored, but it's also used here because the name won't be resolved until type checking is done.

[2]+[3] I did this is in b7d4735, but the result is, that all tests fail for this lint

I've had a look at this, moving the is_argument check to mutate fixes this (only the error that I expected to go is now gone).

[flip1995]

This is the only remaining FN introduced in this PR now. So ping @rust-lang/clippy for review.

[Manishearth]

file an issue for it and refer to it in the test file and we can land tis

[phansch]

@bors r+ p=10

[bors]

📌 Commit 3d39379 has been approved by phansch

[bors]

⌛ Testing commit 3d39379 with merge df80193...

[bors]

☀️ Test successful - checks-travis, status-appveyor
Approved by: phansch
Pushing df80193 to master...

[phansch]

@bors treeclosed-