The stability check checks the `PublicItems` map when giving errors if
there is a #[stable] item with a public contents that doesn't not have
its own stability. Without recording this, struct fields and enum
variants will not get errors for e.g. stable modules with unmarked
functions internally.

This is just improving the compiler's precision to give the standard
library developers more information earlier.

E.g.

    #![staged_api]
    #![feature(staged_api)]
    #![crate_type = "lib"]

    #[stable(feature = "rust1", since = "1.0.0")]
    pub struct Foo {
        pub x: i32
    }

    #[stable(feature = "rust1", since = "1.0.0")]
    pub mod bar {
        pub fn baz() {}
    }

Without the patch it gives:

    test.rs:12:5: 12:20 error: This node does not have a stability attribute
    test.rs:12     pub fn baz() {}
                   ^~~~~~~~~~~~~~~
    error: aborting due to previous error

With the patch it gives:

    test.rs:7:9: 7:15 error: This node does not have a stability attribute
    test.rs:7     pub x: i32
                      ^~~~~~
    test.rs:12:5: 12:20 error: This node does not have a stability attribute
    test.rs:12     pub fn baz() {}
                   ^~~~~~~~~~~~~~~
    error: aborting due to 2 previous errors

We were recording stability attributes applied to fields in the
compiler, and even annotating it in the libs, but the compiler didn't
actually do the checks to give errors/warnings in user crates.

 Turn `unsafe_no_drop_flag` back into a gated-feature.

Fix #22173

 Currently we have a `set_mode` mutator, so this just adds the pairing of a
`mode` accessor to read the value.

Closes #22738

 If the filename for a path is `None` then we know that the creation of the
parent directory created the whole path so there's no need to retry the call to
`create_dir`.

Closes #22737

 Specifically, the following actions were takend:

* The `copy_memory` and `copy_nonoverlapping_memory` functions
  to drop the `_memory` suffix (as it's implied by the functionality). Both
  functions are now marked as `#[stable]`.
* The `set_memory` function was renamed to `write_bytes` and is now stable.
* The `zero_memory` function is now deprecated in favor of `write_bytes`
  directly.
* The `Unique` pointer type is now behind its own feature gate called `unique`
  to facilitate future stabilization.

[breaking-change]

 This commit moves `std::env` away from the `std::old_io` error type as well as
the `std::old_path` module. Methods returning an error now return `io::Error`
and methods consuming or returning paths use `std::path` instead of
`std::old_path`. This commit does not yet mark these APIs as `#[stable]`.

This commit also migrates `std::old_io::TempDir` to `std::fs::TempDir` with
essentially the exact same API. This type was added to interoperate with the new
path API and has its own `tempdir` feature.

Finally, this commit reverts the deprecation of `std::os` APIs returning the old
path API types. This deprecation can come back once the entire `std::old_path`
module is deprecated.

[breaking-change]

 This commit removes many unnecessary `unsafe impl` blocks as well as pushing the
needed implementations to the lowest level possible. I noticed that the bounds
for `RwLock` are a little off when reviewing #22574 and wanted to ensure that we
had our story straight on these implementations.

 r? @Gankro

 The recent commit on liblibc for nacl break the compilation for OpenBSD (and Bitrig too, I think).

The problem is `ino_t` come from another block too now. This patch remove the extra declaration.

```
.../src/liblibc/lib.rs:98:9: 98:37 error: a type named `ino_t` has already been imported in this module [E0251]
.../src/liblibc/lib.rs:98 pub use types::os::arch::posix01::*;
                                  ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
error: aborting due to previous error
```

@dhuseby do you have this problem too ? and if yes, does this patch correct it ?

 Two changes:

1. Make traits with assoc types invariant w/r/t their inputs.
2. Fully normalize parameter environments, including any region variables (which were being overlooked).

The former supports the latter, but also just seems like a reasonably good idea.

Fixes #21750

cc @edwardw
r? @pnkfelix

 According to #22650 and bb0bbf63 the
link changed.
I think this should fix the link on http://doc.rust-lang.org/#faqs

 Rather than stabilize on the current API, we're going to punt this
concern to crates.io, to allow for faster iteration.

[breaking-change]

 r? @steveklabnik

This is a fix for #20300 though as a side-sweep it fixes a number of stack overflows because it integrates cycle detection into the conversion process. I didn't go through and retest everything.

The tricky part here is that in some cases we have to go find the information we need from the AST -- we can't use the converted form of the where-clauses because we often have to handle something like `T::Item` *while converting the where-clauses themselves*. Since this is also not a fixed-point process we can't just try and keep trying to find the best order. So instead I modified the `AstConv` interface to allow you to request the bounds for a type parameter; we'll then do a secondary scan of the where-clauses to figure out what we need. This may create a cycle in some cases, but we have code to catch that.

Another approach that is NOT taken by this PR would be to "convert" `T::Item` into a form that does not specify what trait it's using. This then kind of defers the problem of picking the trait till later. That might be a good idea, but it would make normalization and everything else much harder, so I'm holding off on that (and hoping to find a better way for handling things like `i32::T`).

This PR also removes "most of" the `bounds` struct from `TypeParameterDef`. Still a little ways to go before `ParamBounds` can be removed entirely -- it's used for supertraits, for example (though those really ought to be converted, I think, to a call to `get_type_parameter_bounds` on `Self` from within the trait definition).

cc @jroesch 

Fixes #20300 

Fix #22173

Fixes #22424.

Fixes #22423.

This commit moves `std::env` away from the `std::old_io` error type as well as
the `std::old_path` module. Methods returning an error now return `io::Error`
and methods consuming or returning paths use `std::path` instead of
`std::old_path`. This commit does not yet mark these APIs as `#[stable]`.

This commit also migrates `std::old_io::TempDir` to `std::fs::TempDir` with
essentially the exact same API. This type was added to interoperate with the new
path API and has its own `tempdir` feature.

Finally, this commit reverts the deprecation of `std::os` APIs returning the old
path API types. This deprecation can come back once the entire `std::old_path`
module is deprecated.

[breaking-change]

According to #22650 and bb0bbf63 the
link changed

Specifically, the following actions were taken:

* The `copy_memory` and `copy_nonoverlapping_memory` functions
  to drop the `_memory` suffix (as it's implied by the functionality). Both
  functions are now marked as `#[stable]`.
* The `set_memory` function was renamed to `write_bytes` and is now stable.
* The `zero_memory` function is now deprecated in favor of `write_bytes`
  directly.
* The `Unique` pointer type is now behind its own feature gate called `unique`
  to facilitate future stabilization.
* All type parameters now are `T: ?Sized` wherever possible and new clauses were
  added to the `offset` functions to require that the type is sized.

[breaking-change]

 Apply borrowck to fns that appear in const declarations.
Fixes #22382.

r? @EddyB