* Add rustfmt.toml for 2018 edition fmt

* Part-way there

* Closer to upstream tokio

* No more MyFuture

* Port tests

* More stuff to async fn

* Use ? in tests over unwrap

* Workaround for rust-lang/rust#46415

* All tests pass

* async/await is only on nightly for now

* Only nightly on circle as well

* CI is hard

* Prep for async named pipes

* Don't fail tests if local infiles aren't supported

* No more workaround for taiki-e/pin-project#68

* Attempt at windows support

* PollEvented in tokio_net::util

* Avoid compilation error in Transaction::new

* Fix compilation error in tls::connect_async()

* Fix benches. Add SSL env var for tests.

* Test SSL during CI

* Bump dependencies

135: Move most of the processing to proc-macro-derive r=taiki-e a=taiki-e

To generate the correct `Unpin` implementation and the projection methods, we need to collect the types of the pinned fields.  However, since proc-macro-attribute is applied before `#[cfg]` and `#[cfg_attr]` on fields, we cannot be collecting field types properly at this timing. So instead of generating the `Unpin` implementation and the projection methods here, delegate their processing to proc-macro-derive.

#77 moved only the automatic generation of `Unpin` implementation to proc-macro-derive but found that it could not support `#[cfg_attr]`, so this moves more processing.

See also #68 (comment).

This also adds support for the use of `#[cfg]` on tuple structs and tuple variants.

Closes #123

Co-authored-by: Taiki Endo <[email protected]>

Recently, the getters for the priority field in
`uavcan::session_id::message::MessageSessionId` were disabled from being
generated to avoid an unused warning, as they are, indeed, unused in the build
codebase.

Nonetheless, the getter for `priority` is actually used by one of the tests, such
that is needed, but only when testing.

The general way to do this, would be to add a `#[cfg_attr(...)]` for the `skip`
attribute to avoid generating it during testing.

Unfortunately, `cfg*` attributes are not expanded before `proc_macro_attributes`
expansion and the `modular-bitfield` does not manage them manually.

This is a general problem in the proc-macro environment, with an example being
taiki-e/pin-project#68.

The rust team is working on a solution to this, see, for example:

https://doc.rust-lang.org/beta/unstable-book/library-features/cfg-eval.html
https://doc.rust-lang.org/beta/unstable-book/language-features/macro-attributes-in-derive-output.html

And the related issues.

In particular, `cfg_eval` provides a way to ensure that the `cfg` attributes are
evaluated before the expansion of the following macros.

For this reason, the `cfg_eval` was activated, a `#[cfg_eval]` attribute was
added to `MessageSessionId` and a `cfg_attr` conditioned on `test` was added to
the priority field of the same structure.

As this is not yet stabilized, the crate now depends on nightly, and cannot be
compiled in stable.

It was possible to provide workarounds to avoid moving outside stable.
For example, by duplicating `MessageSessionsId` and providing a version with the
`skip` attribute and one without, conditioning the whole structure, instead of a
field attribute, on `test`.

Nonetheless, being nightly only is not considered a problem for this crate
usage and the "forward-correct" solution was thus preferred.