Analysis of rustc test failures under polonius
===

Listing of the test failures of the polonius compare mode, now that [#60171](https://github.com/rust-lang/rust/pull/60171) has merged. (This PR switched the NLL compare-mode from `-Zborrowck=migrate` to `-Zborrowck=mir`, the polonius compare-mode already being `-Zborrowck=mir -Zpolonius`. Otherwise there'd be 200+ compare-mode failures because of the NLL migrate mode default)

Each test failure will have:
- a link to the test source code
- a link to the NLL/default test expectations from master or PR #60171
- a link to the Polonius test expectations
- a link to a diff between the NLL and Polonius outputs
- a small WIP blurb about the polonius output

when applicable: that is, when the difference is actually meaningful and not to be ignored or `--bless`ed.

My current ideas were:
- ignore the test if the Polonius output is identical to the NLL output: instead of creating yet another level of test revision combinations, at least until we execute Polonius tests in CI (and when Polonius is more advanced), and re-review to add the test expectations then. I also think that if we add manual Polonius revisions now, they'd be executed in CI and not only in the polonius compare-mode. While this is eventually what we'd want, maybe that time isn't _now_.
- for differences in diagnostics only: bless the output. When Polonius and its integration is more advanced, we can review those and file issues specific to each difference. (These differences are more about "completeness" than "correctness", and can/will be the focus at a later stage.)

Summary:
- mostly diagnostics differences
- some error differences in closures (which might require chalk)
- 2 OOMs in fact generation (related to the "`Locations:All` hack") and for which I have a local fix

Notable PRs:
- [#62736](https://github.com/rust-lang/rust/pull/62736) 2019/07/17
- [#64554](https://github.com/rust-lang/rust/pull/64554) 2019/09/17
- [#64749](https://github.com/rust-lang/rust/pull/64749) 2019/09/24 - fixes a lot of the diagnostics differences described below, which were added before july, and the overflow case #24.
- [#70546](https://github.com/rust-lang/rust/pull/70546) 2020/03/30 - fixes the move errors false positives, and some trivial diagnostics wording

---

#### Tests

#### 1. [borrowck/borrowck-escaping-closure-error-2.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/borrowck-escaping-closure-error-2.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/borrowck-escaping-closure-error-2.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/borrowck/borrowck-escaping-closure-error-2.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-borrowck-borrowck-escaping-closure-error-2-diff) :white_check_mark:

Tests the special handling for closures escaping the current frame while holding on to things owned by the function.

NLL returns E0373, and Polonius a regular "borrowed value does not live long enough" E0597.

Category: diagnostics.
Resolution: bless the output, and file an issue about the specific diagnostic.

-> blessed in [this commit](https://github.com/lqd/rust/commit/9b2a6df596a4a52b4fc7e80ecdc6c60ce8414b7c)

---

#### 2. [borrowck/borrowck-lend-flow-loop.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/borrowck-lend-flow-loop.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/borrowck-lend-flow-loop.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/borrowck/borrowck-lend-flow-loop.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-borrowck-borrowck-lend-flow-loop-diff) :white_check_mark:

At first glance, looks like a real NLL/Polonius difference, i.e. a bug (related to loops ?)
The 2 E0502s are the same. Polonius has 2 additional E0499s.

There are other "interesting cases" involving loops: in #8 and #20.

After talking with Matthew: there are problems in fact generation which could be related here (rather than being problems in Polonius per se).

Category: bug in fact generation.
Resolution: As with some of the `rand` cases reductions, this seems like missing `killed` facts. The `Call` terminator doesn't kill the loans of the destination place. `StorageDead` doesn't kill the loans either. I have a fix locally.

-> fixed in [this commit](https://github.com/lqd/rust/commit/cdbbf5f5bc015ce0965ea4ba867b5f2035a6760b)

Note (unrelated to Polonius): [this comment](https://github.com/rust-lang/rust/blob/c21fbfe7e310b9055ed6b7c46b7d37b831a516e3/src/test/ui/borrowck/borrowck-lend-flow-loop.rs#L119) mentions wanting to use `loop {}` but the test does not do that.

---

#### 3. [borrowck/issue-45983.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/issue-45983.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/issue-45983.nll.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/borrowck/issue-45983.migrate.polonius.stderr) :white_check_mark:

This is just a difference from the test construction: it's ignored in compare-mode-nll and manually tests the migrate and NLL modes. The polonius compare-mode is therefore compared against the "migrate" revision but ran with full NLLs.

There is no actual difference in the errors output by NLLs and Polonius.

Category: not an issue.

-> ignored in [this commit](https://github.com/rust-lang/rust/commit/0f23e6bc515a45b14f4dbb44f1ff92cf312aa203).

---

#### 4. [borrowck/promote-ref-mut-in-let-issue-46557.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/promote-ref-mut-in-let-issue-46557.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/promote-ref-mut-in-let-issue-46557.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/borrowck/promote-ref-mut-in-let-issue-46557.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-borrowck-promote-ref-mut-in-let-issue-46557-diff) :white_check_mark:

NLLs and Polonius have similar errors, but NLL seems to have a special cased diagnostic when returning temporary values ("returns a value referencing data owned by the current function" for the returned expression) which Polonius sees as a regular use ("borrow later used here" for the returned expression) ?

Of course, the former being more specific avoids the latter's more generic and inapplicable "note: consider using a `let` binding to create a longer lived value".

In effect, the errors point at different places: NLL's E0515 point to the _return of the temporary_ while Polonius' E0716 points at the creation of the temporary (but also notes with the use/drop). It would be interesting to look at the `polonius-engine` errors directly.

Resolution:`polonius-engine` returns the errors at the `StorageDead` of the temporaries at the end of the scope, which seems correct. So I assume this is just another issue to file about diagnostics under Polonius. It seems ok to bless the output in its current state.

-> blessed in [this commit](https://github.com/lqd/rust/commit/8976d912303795821bfc3a1b4e20b1911962e835)

---

#### 5. [borrowck/two-phase-reservation-sharing-interference-2.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/two-phase-reservation-sharing-interference-2.rs), revisions `migrate2015` and `nll2015` — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/two-phase-reservation-sharing-interference-2.nll2015.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/borrowck/two-phase-reservation-sharing-interference-2.migrate2015.polonius.stderr) :white_check_mark:

This is just a difference from the test construction, it's ignore-compare-mode-nll and manually checks migrate/nll over edition2015/2018.

This failure is because the `migrate2015` revision is ran with `-Zpolonius`. There is no actual difference in the errors output by NLLs and Polonius.

Category: not an issue.
Resolution: ignore this test with Polonius for now.

-> ignored in [this commit](https://github.com/rust-lang/rust/commit/ad14b04863108b07378d77e107b93a65fbf9425e).

---

#### 6. [borrowck/two-phase-reservation-sharing-interference-2.rs](https://github.com/rust-lang/rust/blob/c21fbfe7e310b9055ed6b7c46b7d37b831a516e3/src/test/ui/borrowck/two-phase-reservation-sharing-interference-2.rs), revisions `migrate2018` and `nll2018` — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/two-phase-reservation-sharing-interference-2.nll2018.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/borrowck/two-phase-reservation-sharing-interference-2.migrate2018.polonius.stderr) :white_check_mark:

This is the same as #5, with a different revision: the `migrate2018` revision is ran with `-Zpolonius`, creating this failure. There is no actual difference in the errors output by NLLs and Polonius.

Category: not an issue.
Resolution: same as with failure #5 (♫ A little bit of NLLs in my life. A little bit of Polonius by my side. Failure Number Five ♪ — (c) Lou Bega + lqd 2019)

-> ignored in [this commit](https://github.com/rust-lang/rust/commit/ad14b04863108b07378d77e107b93a65fbf9425e).

---

#### 7. [borrowck/two-phase-surprise-no-conflict.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/two-phase-surprise-no-conflict.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/borrowck/two-phase-surprise-no-conflict.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/borrowck/two-phase-surprise-no-conflict.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-borrowck-two-phase-surprise-no-conflict-diff) :white_check_mark:

Both have the same 15 errors, but have differences in diagnostics/notes:
- the origin of the free region 'a is missing with Polonius
- the source/span of an immutable borrow is different
- and the note seems _a bit better_ with Polonius ?

NLL:

error[E0502]: cannot borrow `*reg` as mutable because it is also borrowed as immutable
  --> $DIR/two-phase-surprise-no-conflict.rs:141:5
   |
LL | fn register_plugins<'a>(mk_reg: impl Fn() -> &'a mut Registry<'a>) {
   |                     -- lifetime `'a` defined here
...
LL |     reg.register_univ(Box::new(CapturePass::new(&reg.sess_mut)));
   |     ^^^^^^^^^^^^^^^^^^-----------------------------------------^
   |     |                 |                         |
   |     |                 |                         immutable borrow occurs here
   |     |                 cast requires that `reg.sess_mut` is borrowed for `'a`
   |     mutable borrow occurs here

Polonius:

error[E0502]: cannot borrow `*reg` as mutable because it is also borrowed as immutable
  --> $DIR/two-phase-surprise-no-conflict.rs:141:5
   |
LL |     reg.register_univ(Box::new(CapturePass::new(&reg.sess_mut)));
   |     ^^^^-------------^^^^^^^^^^^^^^^^^^^^^^^^^^^-------------^^^
   |     |   |                                       |
   |     |   |                                       immutable borrow occurs here
   |     |   immutable borrow later used by call
   |     mutable borrow occurs here

Is this diagnostic special-cased in NLLs for casts as the note mentions ? (is there one cast desugared here ?)

Category: diagnostics.
Resolution: seems ok to bless the polonius output, and file an issue about this specific diagnostic.

-> blessed in [this commit](https://github.com/lqd/rust/commit/594e9bf6585f84a30db1d6e457ce60c038047220)

---

#### 8. [consts/promote_const_let.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/consts/promote_const_let.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/consts/promote_const_let.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/consts/promote_const_let.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-consts-promote_const_let-diff) :white_check_mark:

Same errors, but with a difference in diagnostics about the user-provided type annotations for E0597.

NLL:

LL |     let x: &'static u32 = {
   |            ------------ type annotation requires that `y` is borrowed for `'static`

Polonius:

LL |     let x: &'static u32 = {
   |         - borrow later stored here

Category: diagnostics.
Resolution: seems ok to bless the polonius output, and file an issue about this specific diagnostic.

-> blessed in [this commit](https://github.com/rust-lang/rust/commit/e3251614d614c8c5bb2ec3854e149823aa3c4735).

---

#### 9. [dropck/dropck_trait_cycle_checked.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/dropck/dropck_trait_cycle_checked.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/dropck/dropck_trait_cycle_checked.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/dropck/dropck_trait_cycle_checked.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-dropck-dropck_trait_cycle_checked-diff) :white_check_mark:

Both have the same errors, and are very similar to the "cast requiring 'static borrows" ones from #7, but for E0597 here.

There's a difference in a diagnostic's notes (the "casts" note on `O::new()` are sometimes missing in Polonius).

Polonius sometimes also has notes like `borrow might be used here, when "o1" is dropped and runs the destructor for type "std::boxed::Box<dyn Obj<'_>>"`, is this a special-cased diagnostic about Box and its destructor in NLLs (so that it doesn't appear in the errors) ?

Category: diagnostics.
Resolution: seems ok to bless the polonius output, and file an issue about this specific diagnostic.

-> blessed in [this commit](https://github.com/lqd/rust/commit/4b28e8831e1d4d2f3bab39780c2d2d6644495de5)

---

#### 10. [feature-gates/feature-gate-nll.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/feature-gates/feature-gate-nll.rs) :white_check_mark:

A test about turning the NLL feature gate on, ignored by the NLL compare-mode.

Category: not an issue.
Resolution: ignore this test with Polonius for now.

-> ignored in [this commit](https://github.com/rust-lang/rust/commit/51ecb20faad94b0dc93388bc808dcd75c6e82298).

---

#### 11. [generator/ref-escapes-but-not-over-yield.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/generator/ref-escapes-but-not-over-yield.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/generator/ref-escapes-but-not-over-yield.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/generator/ref-escapes-but-not-over-yield.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-generator-ref-escapes-but-not-over-yield-diff) :white_check_mark:

The NLL error seems special-cased for borrows escaping generators (E0521 instead of Polonius' usual E0597), but the source of error seems the same.

Category: is this just a diagnostics special case, or a difference in behaviour with generators ?
Resolution: seems ok to bless the Polonius output, and file an issue about this specific diagnostic.

-> blessed in [this commit](https://github.com/rust-lang/rust/commit/929ef24dbcffc85c5cb66eef96cb15911e5b7ee0).

---

#### 12. [issues/issue-45696-scribble-on-boxed-borrow.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/issues/issue-45696-scribble-on-boxed-borrow.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/issues/issue-45696-scribble-on-boxed-borrow.nll.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/issues/issue-45696-scribble-on-boxed-borrow.migrate.polonius.stderr) :white_check_mark:

Once again, the difference is in the test construction, it is ignored in compare-mode NLL and tested manually with revisions, and exists because the `migrate` revision is ran with `-Zpolonius`. There is no actual difference in the errors output by NLLs and Polonius.

Category: not an issue.
Resolution: as for the other cases, ignore this in compare-mode polonius (and later test polonius with a manual revision).

-> ignored in [this commit](https://github.com/rust-lang/rust/commit/207adc0ebf062692a15848584ef70cc013018bea).

---

#### 13. [nll/get_default.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/get_default.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/get_default.nll.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/nll/get_default.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-nll-get_default-diff) :white_check_mark:

Polonius has 1 error, NLLs have 3, but those 2 missing look to be "fixed by Polonius".

The 2 fixed NLL errors, tagged "Ideally, this would not error.":

error[E0502]: cannot borrow `*map` as mutable because it is also borrowed as immutable
  --> $DIR/get_default.rs:21:17
   |
LL | fn ok(map: &mut Map) -> &String {
   |            - let's call the lifetime of this reference `'1`
LL |     loop {
LL |         match map.get() {
   |               --- immutable borrow occurs here
LL |             Some(v) => {
LL |                 return v;
   |                        - returning this value requires that `*map` is borrowed for `'1`
...
LL |                 map.set(String::new()); // Ideally, this would not error.
   |                 ^^^ mutable borrow occurs here

error[E0502]: cannot borrow `*map` as mutable because it is also borrowed as immutable
  --> $DIR/get_default.rs:37:17
   |
LL | fn err(map: &mut Map) -> &String {
   |             - let's call the lifetime of this reference `'1`
LL |     loop {
LL |         match map.get() {
   |               --- immutable borrow occurs here
...
LL |                 return v;
   |                        - returning this value requires that `*map` is borrowed for `'1`
...
LL |                 map.set(String::new()); // Ideally, just AST would error here
   |                 ^^^ mutable borrow occurs here

error: aborting due to 3 previous errors

There is one diagnostic difference with Polonius, which misses the "let's call the lifetime of this reference" notes:

LL | fn err(map: &mut Map) -> &String {
   |             - let's call the lifetime of this reference `'1`

Category: diagnostics.
Resolution: seems ok to bless the polonius output, and file an issue about this specific diagnostic.

-> blessed in [this commit](https://github.com/rust-lang/rust/commit/83f3b7bef82e87d35227e751f6db84d8a8ee2b50).

---

#### 14. [nll/issue-46589.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/issue-46589.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/issue-46589.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/nll/issue-46589.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-nll-issue-46589-diff) :white_check_mark:

References [issue #46589 "MIR borrowck doesn't accept the example of iterating and updating a mutable reference](https://github.com/rust-lang/rust/issues/46589)"

The history of this issue is a bit hard to pinpoint at first glance: this initial issue expected to work "in the future", and indeed works today in both NLL and Polonius.

An example was added later as a test, and the error which was removed from the NLL output, and caused this issue to be fixed, is present in the Polonius output.
From a cursory look, it feels related to assignment to projections ? (which we don't handle in fact generation yet, so Polonius might be missing the expected `killed` facts here).

Particularly, this simple deref `bb8[0]: (*_2) = &mut (*_4);` doesn't kill the loans on `_2`.

-> Fixed in [this commit](https://github.com/lqd/rust/commit/aa1b855da4f91028f7f70070d755a5a2d608c824).

The NLL error in the match statement seems expected to be accepted by Polonius (it's borrowing and killing the same variable) ? Or is that another Polonius bug here ?

---

#### 15. [nll/issue-55850.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/issue-55850.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/issue-55850.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/nll/issue-55850.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-nll-issue-55850-diff) :white_check_mark:

Those are the same 2 errors, but in the reverse order.

Category: not an issue.
Resolution: if this order is deterministic, the polonius output can be blessed.
-> However, as of writing this, this test passes now.

---

#### 16. [nll/loan_ends_mid_block_pair.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/loan_ends_mid_block_pair.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/loan_ends_mid_block_pair.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/nll/loan_ends_mid_block_pair.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-nll-loan_ends_mid_block_pair-diff) :white_check_mark:

Polonius finds an error for each of the invalid `data.0 =` assignments, while NLLs maybe only see the 1st one ? [This playground](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=df07d77aff304910fac5613a94cf0b54) seems to validate this hypothesis: commenting the 1st assignment emits only an error for the 2nd one, not the 3rd.

Update: the fact that Polonius emits 3 errors seems like a bug, even if it's interesting. Fixing fact generation of the `killed` relation, especially around assignments to projections, makes Polonius output a single error, like NLL. With this fix the Polonius and NLL output are the same.

Category: not an issue.

-> blessed in [this commit](https://github.com/rust-lang/rust/commit/815e1721c1e5641a0db7bb419cfefab09cf658e3).

---

#### 17. [nll/return-ref-mut-issue-46557.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/return-ref-mut-issue-46557.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/nll/return-ref-mut-issue-46557.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/nll/return-ref-mut-issue-46557.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-nll-return-ref-mut-issue-46557-diff) :white_check_mark:

This one looks like #4: NLLs and Polonius have similar errors, but NLLs seem to have a special cased diagnostic when returning temporary values.

Resolution: same as #4, the output seems to only differ in how diagnostics are done over the Polonius output.

-> blessed in [this commit](https://github.com/lqd/rust/commit/2814e30ef2730c371deee37fbedf5e0bcbdedbe2)

---

#### 18. [span/regions-escape-loop-via-vec.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/span/regions-escape-loop-via-vec.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/span/regions-escape-loop-via-vec.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/span/regions-escape-loop-via-vec.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-span-regions-escape-loop-via-vec-diff) :white_check_mark:

Looks like an "interesting difference". Polonius has one more error, about the assignment to z, which is borrowed in earlier iterations of the loop. Could be related to #2.

Polonius:

error[E0506]: cannot assign to `z` because it is borrowed
  --> $DIR/regions-escape-loop-via-vec.rs:6:21
   |
LL |         let mut z = x;
   |                     ^ assignment to borrowed `z` occurs here
LL |         _y.push(&mut z);
   |         --      ------ borrow of `z` occurs here
   |         |
   |         borrow later used here

After talking with Matthew: there are problems in fact generation which could be related here (rather than being problems in Polonius per se).

Category: bug in fact generation.
Resolution: As with some of the `rand` cases reductions, this seems like missing `killed` facts. The `Call` terminator doesn't kill the loans of the destination place. `StorageDead` doesn't kill the loans either.

-> fixed in [this commit](https://github.com/lqd/rust/commit/7dd1b76efc6b3b499da3902113e36fb6d2166092)

---

#### 19. [unboxed-closures/unboxed-closures-failed-recursive-fn-1.rs](https://github.com/rust-lang/rust/blob/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/unboxed-closures/unboxed-closures-failed-recursive-fn-1.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/8fa44c061b7d2db63eed229810f34f1e13633eea/src/test/ui/unboxed-closures/unboxed-closures-failed-recursive-fn-1.stderr), [Polonius](https://raw.githubusercontent.com/lqd/rust/c40be9633d1eef5cd89e8fc934b1c62de3493a59/src/test/ui/unboxed-closures/unboxed-closures-failed-recursive-fn-1.polonius.stderr), [diff](https://gist.github.com/lqd/9a81a85b287ef438798754fd1afca445#file-unboxed-closures-unboxed-closures-failed-recursive-fn-1-diff) :white_check_mark:

Looks like a mix between #8, with the diagnostics difference about user-provided type annotations, and #9, with the possible Box special-casing.

There is also another diagnostic difference in the last error (E0506), with

NLL:

LL |     factorial = Some(Box::new(f));
   |     ^^^^^^^^^ assignment to borrowed `factorial` occurs here

Polonius:

LL |     factorial = Some(Box::new(f));
   |     ^^^^^^^^^
   |     |
   |     assignment to borrowed `factorial` occurs here
   |     borrow later used here

in that Polonius displays a "... here" note twice.

Resolution: the errors are at the same location, and only the diagnostics differ. Seems ok to bless the polonius output, and file an issue about this specific diagnostic.

-> blessed in [this commit](https://github.com/lqd/rust/commit/b1c6bca707caa27ebe1e9dc2bf2db3fdbb6badab).

---

#### 20. the `rand` crate difference (not a rustc test but something that needs to be done anyway) :white_check_mark:

There is a difference between NLLs and Polonius, found in `rand 0.4.6` ([extraction](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=7369c3d90d9b0a2ada3a290bf968e876), [reduction](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=f60c7f2cba8a0a144d18bfb79a1b2d38)) which Polonius rejects and NLLs accept.

This once again involves loops, like #2 (which also involves the function's parameters like here) and #18.

-> looks very much like issue-47680 but surprisingly Polonius doesn't like it.

After talking with Matthew: there are problems in fact generation which could be related here (rather than being problems in Polonius per se).

Category: bug in fact generation.
Resolution: The `Call` terminator doesn't kill the loans of the destination place. `StorageDead` doesn't kill the loans either.

-> fixed in [this commit](https://github.com/lqd/rust/commit/cdbbf5f5bc015ce0965ea4ba867b5f2035a6760b)

---

#### 21. [borrowck/borrowck-migrate-to-nll.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/borrowck/borrowck-migrate-to-nll.rs) :white_check_mark:

(Added on 2019/06/11)

A test of the NLL migrate mode, ignored in NLL compare mode.

Category: not an issue.

-> ignored in [this commit](https://github.com/lqd/rust/commit/a72a3b3e1896e2a148915c7ef2d0be116bdfda9d).

---

#### 22. [emit-artifact-notifications.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/emit-artifact-notifications.rs) :white_check_mark:

(Added on 2019/06/11)

This just lacks the expected json output for `-Z polonius`.

Category: not an issue
Resolution: bless the output.

-> blessed in [this commit](https://github.com/rust-lang/rust/commit/6b99b95aa62cbab3c11c42f4707998852ed53d85).

---

#### 23. [save-analysis/emit-notifications.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/save-analysis/emit-notifications.rs) :white_check_mark:

(Added on 2019/06/11)

Same as #22, this just lacks the expected json output for `-Z polonius`.

Category: not an issue
Resolution: bless the output.

-> blessed in [this commit](https://github.com/rust-lang/rust/commit/dacbf9fa6ccea84e63a77f679561db56bc3badb3).

---

#### 24. [issues/issue-38591.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/issues/issue-38591.rs) — output from [Polonius](https://gist.githubusercontent.com/lqd/8407f6a9ce9625d5105e51fd0f297d8c/raw/a1ce6874d996ea2534f519bc54a961aa301c8e17/issue-38591.txt) :white_check_mark:

(Added on 2019/06/26)

References the [issue #38591 "rustc overflows its stack when type checking ~~a simple program~~ polymorphmic recursion"](https://github.com/rust-lang/rust/issues/38591)

A case of polymorphic recursion, which passes with NLL, but fails with Polonius with an overflow.

This error also occurs without Polonius when asking to dump the NLL facts, so it seems to be a bug triggered by the fact generation itself.

After talking with Matthew: this can happen easily, and requires a bit more investigation to know if it'd be easy or hard to fix.

Investigation notes:
- [`compute_drop_data`](https://github.com/rust-lang/rust/blob/6b468c6360cba002ba1e97b70cf51223a498442e/src/librustc_mir/borrow_check/nll/type_check/liveness/trace.rs#L574-L585) runs the [DropckOutlives](https://github.com/rust-lang/rust/blob/9606f6fa64926a84d82e3c62dbdc57f5c10f756d/src/librustc/traits/query/type_op/outlives.rs#L8) query
- reporting the overflow error itself comes from [here](https://github.com/rust-lang/rust/blob/6b468c6360cba002ba1e97b70cf51223a498442e/src/librustc_mir/borrow_check/nll/type_check/liveness/trace.rs#L519-L523)

Why does it only overflow with nll-facts / polonius ?

Fixed in #64749.

---

#### 25. [hrtb/issue-30786.rs](https://github.com/rust-lang/rust/blob/fa0809d3cdacae8638fd7beea2e85310902e21d3/src/test/ui/hrtb/issue-30786.rs) :white_check_mark:

(Added on 2019/07/15)

This is just a difference from the test construction: it’s ignored in compare-mode-nll and manually tests the migrate and NLL modes. The polonius compare-mode is therefore compared against the “migrate” revision but ran with full NLLs.

There is no actual difference in the errors output by NLLs and Polonius.

Category: not an issue.

-> ignored in [this commit](https://github.com/lqd/rust/commit/c82422768637adb8df71c27e5a65eeb26cbb8000).

---

At this point the `run-pass` test suite was merged into the `ui` test suite.

---

#### 26. [async-await/async-borrowck-escaping-closure-error.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/async-await/async-borrowck-escaping-closure-error.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/async-await/async-borrowck-escaping-closure-error.stderr), [Polonius](https://gist.githubusercontent.com/lqd/5150bf0f38acca441b4568ea8fd3eff4/raw/d0be5aa7cacf35471c2c179adc4498cd76d29135/async-borrowck-escaping-closure-error.polonius.stderr), [diff](https://gist.github.com/lqd/48f9b1f34352b6d00722dee56208043d) :white_check_mark:

(Added on 2019/09/16)

A new test from the recent async/await test push. Does not have polonius expectations.

This looks like an instance of case #1 but in an async closure. NLL returns E0373, and Polonius a regular “borrowed value does not live long enough” E0597.

Category: diagnostics.
Resolution: bless the output, and deal with this once we're looking at diagnostics.

-> Blessed in [this commit](https://github.com/lqd/rust/commit/9f4351d406fe904a99a7588ebb1411e369ba71f7).

---

#### 27. [borrowck/return-local-binding-from-desugaring.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/borrowck/return-local-binding-from-desugaring.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/borrowck/return-local-binding-from-desugaring.stderr), [Polonius](https://gist.githubusercontent.com/lqd/152be6291111f87afc678b0176adde8c/raw/dfcc0b53e47ea527c711c35360238c1162b4c0b2/return-local-binding-from-desugaring.polonius.stderr), [diff](https://gist.github.com/lqd/152be6291111f87afc678b0176adde8c#file-return-local-binding-from-desugaring-stderr-diff) :white_check_mark:

(Added on 2019/09/16)

This looks like an instance of case #4, a special cased diagnostic when returning temporary values.

Category: diagnostics.
Resolution: bless the output, and deal with this once we're looking at diagnostics.

-> blessed in [this commit](https://github.com/lqd/rust/commit/222e9201511bb9a6eaca7fc2c0ee0b85d78c4fd7)

---

#### 28. [dropck/dropck_trait_cycle_checked.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/dropck/dropck_trait_cycle_checked.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/dropck/dropck_trait_cycle_checked.stderr), [Polonius](https://gist.githubusercontent.com/lqd/592f85f99d67bab95ac91acd0af1df19/raw/bb0e9a646b43894e8b43cb9760eaf5bfa6b13c14/dropck_trait_cycle_checked.polonius.stderr), [diff](https://gist.github.com/lqd/592f85f99d67bab95ac91acd0af1df19#file-dropck_trait_cycle_checked-stderr-diff) :white_check_mark:

(Added on 2019/09/16)

This test was handled before as case #9. Here's the diff over the existing [.polonius.stderr expectation](https://gist.github.com/lqd/592f85f99d67bab95ac91acd0af1df19#file-dropck_trait_cycle_checked-polonius-stderr-diff).

Looks like some diagnostics differences, and especially some different ordering or naming in the notes, which was changed in [this commit](https://github.com/rust-lang/rust/commit/2ff337a8e286a5b472f71b3bbdc3d4b6b840870f#diff-bd3f80b956148a5d1567aa8698b8a507) and looks wrong.

Category: diagnostics.
Resolution: bless the output, and deal with this once we're looking at diagnostics.

-> blessed in [this commit](https://github.com/lqd/rust/commit/3ef980aeaae3ae92c1ed3bdbefe35c6175ed24fb)

---

#### 29. [json-options.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/json-options.rs) :white_check_mark:

(Added on 2019/09/16)

This just lacks the expected json output for `-Z polonius`.

Category: not an issue
Resolution: bless the output.

-> blessed in [this commit](https://github.com/lqd/rust/commit/f9c73293e231f1179c426311bd7ab11f46473e9a)

---

#### 30. [json-multiple.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/json-multiple.rs) :white_check_mark:

(Added on 2019/09/16)

This just lacks the expected json output for `-Z polonius`.

Category: not an issue
Resolution: bless the output.

-> blessed in [this commit](https://github.com/lqd/rust/commit/34d31673684a34b3858ba8dbc05fff8804a50508)

---

#### 31. [wrapping-int-combinations.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/wrapping-int-combinations.rs)

(Added on 2019/09/16)

Recent test from [#63692](https://github.com/rust-lang/rust/pull/63692)

OOM.

memory allocation of 33550368768 bytes failed

Looks similar to the one in case 32.

450 MB NLL MIR.

130752 outlives constraints with no specific location x 170646 points in the CFG.

36337 "live" locals; 4610 regions outliving free regions.

I think in both cases, it's the many variables created by assert_eq! and panic! whose outlives constraints are being materialized. Replacing them with `assert!` compiles without OOMs (but slow-ish, 8s on my machine while NLL is around 1s). The NLL MIR dump is 100MB smaller.

---

#### 32. [numbers-arithmetic/saturating-float-casts.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/numbers-arithmetic/saturating-float-casts.rs)

(Added on 2019/09/16)

This used to be in the `run-pass` and is now in the `ui` test suite (and so, was missing from this write-up as it focused on the latter).

The OOM is about materializing a lot of outlives constraints. It's possible this big number is caused by the consts/statics in this test.

10948 outlives constraints x 136664 points in the CFG are materialized [here](https://github.com/rust-lang/rust/blob/8301de16dafc81a3b5d94aa0707ad83bdb56a599/src/librustc_mir/borrow_check/nll/type_check/mod.rs#L229-L233).

(After changing `assert_eq!` to `assert!` here again, it compiles fast.)

This function contains 161875 regions, of which 391 directly outlive a free region, and 2737 regions do _transitively_.

---

#### 33. [hrtb/due-to-where-clause.rs](https://github.com/rust-lang/rust/blob/0a985f2c86add139d880882c968e68747b366be6/src/test/ui/hrtb/due-to-where-clause.rs) :white_check_mark:

(Added on 2019/09/20)

This diagnostics test compiles with NLLs and Polonius, so it's marked `ignore-compare-mode-nll` and should be ignored with Polonius mode.

Category: not a Polonius issue (but an NLL one).
Fixed by #64749 for both NLL and Polonius

---

#### 34. [hrtb/hrtb-perfect-forwarding.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/hrtb/hrtb-perfect-forwarding.rs)

(Added on 2020/02/28)

This was blessed in [#67016](https://github.com/rust-lang/rust/pull/67016) where I've mentioned this was because I believe the difference to be related to general closures handling that would require higher-rankedness help from chalk.

Category: waiting for chalk ?!

Since [#69482](https://github.com/rust-lang/rust/pull/69482), there's an additional diagnostics difference: a trivial diagnostics formatting change: [diff for NLL output](https://github.com/rust-lang/rust/commit/b82cd9f6391df865551bc6c756cc29a7993e39be#diff-8d9af1cfe313a0b36a5538b0ede48e16).

Category: not an issue per se.
Resolution: bless the new output

-> blessed in [this commit](https://github.com/lqd/rust/commit/2a7644746b5c5cc7fe214a61fb443efceac06dc5)

---

#### 35. [nll/outlives-suggestion-simple.rs](https://github.com/rust-lang/rust/blob/9434d6b67fb0817e7db5217e6355cbf4c7e402f6/src/test/ui/nll/outlives-suggestion-simple.rs)

(Added on 2020/02/28)

A couple of trivial changes which should be blessed:
- a grammar and a wording change: [diff for NLL output](https://github.com/rust-lang/rust/commit/9434d6b67fb0817e7db5217e6355cbf4c7e402f6#diff-b95aedcb92cdef60e9d813f23ac9281a)

- an interesting formatting change in universal regions relation suggestions:

fn foo4<'a, 'b, 'c>(x: &'a usize) -> (&'b usize, &'c usize) {
    // FIXME: ideally, we suggest 'a: 'b + 'c, but as of today (may 04, 2019), the null error
    // reporting stops after the first error in a MIR def so as not to produce too many errors, so
    // currently we only report 'a: 'b. The user would then re-run and get another error.
    (x, x) //~ERROR lifetime may not live long enough
}

As mentioned in the FIXME, under NLL a [single suggestion is offered](https://github.com/rust-lang/rust/blob/master/src/test/ui/nll/outlives-suggestion-simple.stderr#L49-L60):

= help: consider adding the following bound: `'a: 'b`

while under Polonius 2 errors with [both `'a: 'b` and `'a: 'c` are proposed](https://github.com/rust-lang/rust/blob/9434d6b67fb0817e7db5217e6355cbf4c7e402f6/src/test/ui/nll/outlives-suggestion-simple.polonius.stderr#L49-L73).

This test failed because of a new suggestion in the second error (I think), which combines both suggestions:

help: add bound `'a: 'b + 'c`

that the FIXME mentioned. (The redundant suggestions are still present, as shown below. But overall I believe this is just due to the outlives suggestion mechanism not handling this case correctly yet)

error: lifetime may not live long enough
  --> /home/lqd/work/rust/rust/src/test/ui/nll/outlives-suggestion-simple.rs:22:5
   |
LL | fn foo4<'a, 'b, 'c>(x: &'a usize) -> (&'b usize, &'c usize) {
   |         --  -- lifetime `'b` defined here
   |         |
   |         lifetime `'a` defined here
...
LL |     (x, x) //~ERROR lifetime may not live long enough
   |     ^^^^^^ returning this value requires that `'a` must outlive `'b`
   |
   = help: consider adding the following bound: `'a: 'b`

error: lifetime may not live long enough
  --> /home/lqd/work/rust/rust/src/test/ui/nll/outlives-suggestion-simple.rs:22:5
   |
LL | fn foo4<'a, 'b, 'c>(x: &'a usize) -> (&'b usize, &'c usize) {
   |         --      -- lifetime `'c` defined here
   |         |
   |         lifetime `'a` defined here
...
LL |     (x, x) //~ERROR lifetime may not live long enough
   |     ^^^^^^ returning this value requires that `'a` must outlive `'c`
   |
   = help: consider adding the following bound: `'a: 'c`

help: add bound `'a: 'b + 'c`

Category: not an issue.
Resolution: bless the new output.

-> blessed in [this commit](https://github.com/lqd/rust/commit/c73d5db21fa89c33064724219e1dd5f168449d55)

---

#### 36. [nll/maybe-initialized-drop-implicit-fragment-drop.rs](https://github.com/rust-lang/rust/blob/be5fe051a843cfbbc1ba4fcd347b641417181b8f/src/test/ui/nll/maybe-initialized-drop-implicit-fragment-drop.rs)
#### 37. [nll/maybe-initialized-drop-with-uninitialized-fragments.rs](https://github.com/rust-lang/rust/blob/2a663555ddf36f6b041445894a8c175cd1bc718c/src/test/ui/nll/maybe-initialized-drop-with-uninitialized-fragments.rs)

(Added on 2020/02/28)

Those 2 changes stem from completing move errors in polonius 0.12, and using it in rustc from [#69482](https://github.com/rust-lang/rust/pull/69482). They are test failures as Polonius sees no errors in the code, and thus fail with `error: ui test compiled successfully!`.

Both tests contain a FIXME about the error, [1](https://github.com/rust-lang/rust/blob/be5fe051a843cfbbc1ba4fcd347b641417181b8f/src/test/ui/nll/maybe-initialized-drop-implicit-fragment-drop.rs#L18) and [2](https://github.com/rust-lang/rust/blob/2a663555ddf36f6b041445894a8c175cd1bc718c/src/test/ui/nll/maybe-initialized-drop-with-uninitialized-fragments.rs#L23), and Polonius is both fixing those here, but introducing a new one in case #38.

Category: not an issue iff #38 is fixed, otherwise move errors and borrowcking are stepping on each other's toes and it's a bug.
Resolution: polonius bug, move error false positive

-> fixed in `polonius-engine` 0.12.1 and [this commit](https://github.com/lqd/rust/commit/860f71f141baf8228a9d71a75092ff2378bfee42) in fact generation

---
#### 38. [nll/maybe-initialized-drop-with-fragment.rs](https://github.com/rust-lang/rust/blob/2a663555ddf36f6b041445894a8c175cd1bc718c/src/test/ui/nll/maybe-initialized-drop-with-fragment.rs)

(Added on 20/02/28)

Same origin as #36 and #37, however the expected borrowck error is not emitted.

Category: polonius bug
Resolution: polonius bug, move error false positive

-> fixed in `polonius-engine` 0.12.1 and [this commit](https://github.com/lqd/rust/commit/860f71f141baf8228a9d71a75092ff2378bfee42) in fact generation

---
#### 39. [closures/closure-expected-type/expect-region-supply-region.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/closures/closure-expected-type/expect-region-supply-region.rs)

(Added on 20/02/28)

Same grammar change found in #35, [diff for NLL output](https://github.com/rust-lang/rust/commit/45832839087da140eeb2a85a8b98927ec27ba21c#diff-a676673241ba88a89c76e77e981adfa8).

Category: not an issue.
Resolution: bless the updated grammar output.

-> blessed in [this commit](https://github.com/lqd/rust/commit/fcd12bd2f42202bf75f89c3712d25295af47e1ab)

---
#### 40. [impl-trait/multiple-lifetimes/error-handling.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/impl-trait/multiple-lifetimes/error-handling.rs)

(Added on 20/02/28)

Some `impl Trait` fixes from [#67681](https://github.com/rust-lang/rust/pull/67681): better locating the cause of a universal region error with a user annotation -- [diff for NLL output](https://github.com/rust-lang/rust/pull/67681/files#diff-92a5bbbc2e978f989064dfb1f9be036f).

Equally applicable to Polonius.

Category: not an issue.
Resolution: bless the updated output.

-> blessed in [this commit](https://github.com/lqd/rust/commit/82424634a389a88584ef40008d613695ff7e72ce)

---
#### 41. [nll/user-annotations/closure-substs.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/impl-trait/multiple-lifetimes/error-handling.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/nll/user-annotations/closure-substs.stderr), [Polonius](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/nll/user-annotations/closure-substs.polonius.stderr)

(Added on 20/02/28)

Like #34, this was blessed in [#67016](https://github.com/rust-lang/rust/pull/67016) where I mentioned that this test has some diagnostics differences due to:
- what I believe to be related to general closures handling that would require higher-rankedness help from chalk
- what I believe to be related about how `'static` is handled in Polonius, and how it differs in how NLL produces related errors (possibly creating duplicate errors here) -- and intertwined in this test with closures.

Category: probably warrants a second opinion and/or more detailed investigation.

Since [#69482](https://github.com/rust-lang/rust/pull/69482), there's an additional diagnostics difference: this is a trivial wording change from #35 and #39. The NLL output didn't change, because the new wording is only present [in the polonius errors](https://github.com/rust-lang/rust/blob/master/src/test/ui/nll/user-annotations/closure-substs.polonius.stderr#L53):

50        --> $DIR/closure-substs.rs:29:9
51         |
52      LL |     |x: &i32, b: fn(&'static i32)| {
-          |      -        - `b` is declared here, outside of the closure body
+          |      -        - `b` declared here, outside of the closure body
54         |      |
55         |      `x` is a reference that is only valid in the closure body
56      LL |         b(x);

Category: not an issue per se.
Resolution: bless the updated output.

-> blessed in [this commit](https://github.com/lqd/rust/commit/5af11d261d38c6dad4475c933ffa59a388979cfb)

---
#### 42. [closures/closure-expected-type/expect-region-supply-region-2.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/closures/closure-expected-type/expect-region-supply-region-2.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/closures/closure-expected-type/expect-region-supply-region-2.nll.stderr), [Polonius](https://gist.github.com/lqd/fe6040db809ea5067341945a0c3aa8d1#file-expect-region-supply-region-2-polonius-stderr)

(Added on 20/08/06)

A recent test addition, not blessed yet.
Polonius sees one more error than NLL, which is [somewhat expected](https://github.com/rust-lang/rust/blob/3cfc7fe78eccc754b16981704a098d7bd520e2fd/src/test/ui/closures/closure-expected-type/expect-region-supply-region-2.rs#L18-L20):

Borrowck doesn't get a chance to run, but if it did it should error here.

Category: not an issue.
Resolution: bless the output

---
#### 43. [closures/closure-expected-type/expect-region-supply-region.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/closures/closure-expected-type/expect-region-supply-region.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/closures/closure-expected-type/expect-region-supply-region.stderr), [Polonius](https://gist.github.com/lqd/fe6040db809ea5067341945a0c3aa8d1#file-expect-region-supply-region-polonius-stderr)

(Added on 20/08/06)

Already blessed test. The NLL test output changed with [PR #72362](https://github.com/rust-lang/rust/pull/72362) removing `ReScope`, and the new Polonius output matches it.

Category: not an issue.
Resolution: bless the output

---
#### 44. [hrtb/hrtb-just-for-static.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/hrtb/hrtb-just-for-static.rs) — outputs from [NLL](https://github.com/rust-lang/rust/raw/master/src/test/ui/hrtb/hrtb-just-for-static.nll.stderr), [Polonius](https://gist.github.com/lqd/fe6040db809ea5067341945a0c3aa8d1#file-hrtb-just-for-static-polonius-stderr)

(Added on 20/08/06)

A recent NLL type-checker bug fix from [PR #72493](https://github.com/rust-lang/rust/pull/72493) changed the test output for NLLs, in both migrate mode and full NLLs.

Full NLLs see 3 errors where the migrate mode sees 2, and Polonius only sees the same 2.

The error it misses is:

error: lifetime may not live long enough
  --> $DIR/hrtb-just-for-static.rs:30:5
   |
LL | fn give_some<'a>() {
   |              -- lifetime `'a` defined here
LL |     want_hrtb::<&'a u32>()
   |     ^^^^^^^^^^^^^^^^^^^^ requires that `'a` must outlive `'static`
   |
   = help: consider replacing `'a` with `'static`

for [this function](https://github.com/rust-lang/rust/blob/3cfc7fe78eccc754b16981704a098d7bd520e2fd/src/test/ui/hrtb/hrtb-just-for-static.rs#L27-L31).

This looks like a case requiring Chalk integration to be able to correctly detect the error.

Category: Probably wait for chalk ?
Resolution: Bless the output until then ?

---
#### 45. [hrtb/hrtb-perfect-forwarding.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/hrtb/hrtb-perfect-forwarding.rs) — outputs from [NLL](https://github.com/rust-lang/rust/raw/master/src/test/ui/hrtb/hrtb-perfect-forwarding.nll.stderr), [Polonius](https://gist.github.com/lqd/fe6040db809ea5067341945a0c3aa8d1#file-hrtb-perfect-forwarding-polonius-stderr)

(Added on 20/08/06)

Like previously mentioned, and also in #43, Polonius doesn't see the HRTB error. The test fails only because rustc's test output now contains the number of emitted warnings since [PR #69926](https://github.com/rust-lang/rust/pull/69926):

-error: aborting due to previous error
+error: aborting due to previous error; 4 warnings emitted

Category: Not an issue
Resolution: Bless the output

---
#### 46. [issues/issue-26217.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/issues/issue-26217.rs) — output from [NLL](https://github.com/rust-lang/rust/raw/master/src/test/ui/issues/issue-26217.nll.stderr)

(Added on 20/08/06)

This test fails because it involves an HRTB like the previous 2, and Polonius doesn't see the error without Chalk, accepting the code.

I do not understand why this test started to fail, however: it is older than the previous PRs updating Polonius test expectations and should have failed then.

Category: Likely requires chalk

---
#### 47. [issues/issue-74564-if-expr-stack-overflow.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/issues/issue-74564-if-expr-stack-overflow.rs)

(Added on 20/08/06)

A recent test with a very big MIR if expression, causing an infinite loop or OOM somewhere.
It doesn't occur when just triggering fact generation, which does complete on this test, so it looks to be somewhere in polonius.

Fact generation is however very slow.

Category: Very likely a Polonius bug.
Resolution: investigate precisely which parts of the process cause this issue.

---
#### 48. [lub-glb/old-lub-glb-hr-noteq2.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/lub-glb/old-lub-glb-hr-noteq2.rs)

(Added on 20/08/06)

Recent test from the leak-check PR. Ignored in NLL compare-mode because of [#73154](https://github.com/rust-lang/rust/issues/73154). The full NLL type-checker currently accepts this case of higher-ranked functions, while the old non-NLL type-checker did not. Polonius of course accepts it as well as it doesn't deal with higher-rankedness.

Category: An issue either in NLL solely, or needing both NLL changes and chalk integration
Resolution: ignore the test in the polonius compare mode

---
#### 49. [regions/region-multiple-lifetime-bounds-on-fns-where-clause.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/regions/region-multiple-lifetime-bounds-on-fns-where-clause.rs) — outputs from [NLL](https://raw.githubusercontent.com/rust-lang/rust/master/src/test/ui/regions/region-multiple-lifetime-bounds-on-fns-where-clause.nll.stderr), [Polonius](https://gist.github.com/lqd/fe6040db809ea5067341945a0c3aa8d1#file-region-multiple-lifetime-bounds-on-fns-where-clause-polonius-stderr)

(Added on 20/08/06)

Polonius sees 7 errors where NLL sees 5.

The first one is:

error: lifetime may not live long enough
  --> $DIR/region-multiple-lifetime-bounds-on-fns-where-clause.rs:10:5
   |
LL | fn b<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
   |          --  -- lifetime `'c` defined here
   |          |
   |          lifetime `'b` defined here
...
LL |     *z = *y;
   |     ^^^^^^^ assignment requires that `'b` must outlive `'c`
   |
   = help: consider adding the following bound: `'b: 'c`

help: add bound `'b: 'a + 'c`

On this function

fn b<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
    // Illegal now because there is no `'b:'a` declaration.
    *x = *y; //~ ERROR E0623
    *z = *y; //~ ERROR E0623
}

the migrate mode sees the same 2 errors as Polonius, while full NLLs only see one. Since the test expectations require an error on line 10 I'm confused as to how this test passes. Probably the annotations are not checked in compare-mode and only the blessed output is compared.

This looks like a case of full NLLs stopping at the firt error, rather than missing errors because of a bug. So the polonius behaviour here looks correct.

For the 2nd additional error, things are similar: on this function

fn c<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) {
    // Here we try to call `foo` but do not know that `'a` and `'b` are
    // related as required.
    a(x, y, z); //~ ERROR lifetime mismatch [E0623]
}

with `fn a<'a, 'b, 'c>(x: &mut &'a isize, y: &mut &'b isize, z: &mut &'c isize) where 'b: 'a + 'c`.

Polonius sees the 2 expected subset errors to NLL's seemingly stopping at the 1st.

Category: Not an issue
Resolution: Bless the output