Add DedupGroupAnnotation phase

Remove the unused install-circt composite GitHub action now that Chisel
has switched to using circt/install-circt.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

Switch from the internal composite GitHub action "install-circt" to a new
one that is published by CIRCT.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

Nightly CIRCT binary builds are taking too long due to the full library
build.  Move the Chisel job back by one hour to give it more slack.  On
2023/10/05, the job started 24 minutes too early.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

Co-authored-by: Jack Koenig <koenig@sifive.com>

Co-authored-by: Jiuyang Liu <liu@jiuyang.me>
Co-authored-by: Megan Wachs <megan@sifive.com>
Co-authored-by: Jack Koenig <koenig@sifive.com>

Fix and test probe of ChiselEnum.

Add a requirement check that Optional Group definitions are nested in the
same way that their declarations are.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

Change the Builder to record the Root Optional Group in its Group Stack.
This is done to later enable error checking that group definitions are
nested correctly.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

While the fix avoids the Chisel error encountered when trying to rwprobe
an enum, actually there are still enum annotations emitted for the port!

This is not a problem in practice, however, as firtool ignores them
without trying to resolve them (which would fail).

Add support for optional groups.  These are groups of statements with
optional functionality.  The primary use case is intended for verification
code to "extract" such code into bound modules.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

Don't add enum annotations to probes of enum.

- Provide a public simpleClassName utility to generate simple class names in a non-error prone way, particularly when generating inner class names. All typeNames are modified to now use this utility.
- Mixing HasAutoTypename into anonymous bundles is now prohibited by the compiler, as this inevitably leads to name conflicts.

Change the reported version of Chisel-emitted FIRRTL to 3.3.0.  Version
3.2.0 includes groups and Chisel is currently emitting things that are not
in a spec version yet, e.g., property lists.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

fmt

Add tests of FixedIORawModule and friends.  Rename the file containing
FixedIORawModule to "FixedIOModule".  This previously used the old
"Polymorphic" name.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

Add a new module hierarchy that promotes the type of the IO of that module
to a type parameter.  This module hierarchy includes:

  1. FixedIOBaseModule
  2. FixedIORawModule
  3. FixedIOBlackBox

Both (2) and (3) inherit from (1).  These modules are all "fixed" in the
type of their IO.  It follows that no more IO are allowed to be created
other than what is in the type.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

Add a `private[chisel3]` ability to block further IO creation in a
`BaseModule`.  This is intended for advanced, in-tree APIs that need to
disallow the creation of further IO by a user.  This API has mutable state
to track if more IO is allowed to be created, a method to get the
value, and a method to set the module to disallow further IO creation.
This one-way setting (h/t @jackkoenig) tries to keep this as locked down
as possible.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

* MixedVec is a Seq
* Add a test for MixedVec being a Seq
* fix integration test -- note that this same change would have been needed for VecInit(Vec(...))

Change `FlatIO` to work for any `Data` as opposed to just for `Record`.
For non-`Record`, have this behave the same as `IO`.  I.e., there is no
prefix stripping.

Signed-off-by: Schuyler Eldridge <schuyler.eldridge@sifive.com>

This improves error messages involving views where the view maps to a
single Data. The error messages on legacy connections (:= and <>)
involving views that do not correspond 1-1 with a single target Data are
still bad, but connectables (eg. :#=) do benefit even in that case.