Phase 13.6 static channels

[JustinKovacich]

Introduce a static-pool ChannelFactory

[Copilot]

Pull request overview

Introduces a static-pool, bare-metal–friendly ChannelFactory backend and updates the transport/channel abstraction so channel construction can be backed by per-type pools (enabling “no heap after Client::new” for channel storage).

Changes:

• Added static_channels module providing OneshotPool/MpscPool primitives plus the define_static_channels! macro for generating per-T pooled channel wiring.
• Reshaped transport::ChannelFactory to delegate constructors via OneshotPooled/BoundedPooled/UnboundedPooled and made bounded channel associated types capacity-typed (const N).
• Updated client internals, embassy/tokio backends, and added witness tests for bare-metal/static-pool integration and allocation behavior.

Reviewed changes

Copilot reviewed 16 out of 17 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
src/static_channels/mod.rs	New static-pool channel primitives and define_static_channels! macro.
src/transport.rs	Updates ChannelFactory to use pooled traits + const-capacity bounded types; adds pooled trait definitions/docs.
src/tokio_transport.rs	Adapts TokioChannels to new pooled-trait + const-capacity Bounded* associated types.
src/embassy_channels.rs	Adapts EmbassySyncChannels to new pooled-trait + const-capacity Bounded* associated types.
src/client/mod.rs	Exposes message types for static pool declarations; adds required pooled bounds.
src/client/inner.rs	Makes ControlMessage public and adds pooled bounds required by new factory shape.
src/client/socket_manager.rs	Updates bounded channel types to ..., 16 and adds pooled bounds needed for construction.
src/lib.rs	Exposes static_channels module under bare_metal; minor re-export cleanup.
tests/bare_metal_client.rs	Switches witness test to macro-declared static channels backend.
tests/static_channels_alloc_witness.rs	New integration test using a counting allocator to witness “no allocations” for static-pool hot-path ops.
Cargo.toml	Registers new static_channels_alloc_witness integration test.
src/server/subscription_manager.rs	Formatting-only change.
src/server/mod.rs	Formatting-only test change.
examples/discovery_client/src/main.rs	Formatting-only change.
examples/client_server/src/main.rs	Formatting-only change.
examples/bare_metal/src/main.rs	Formatting-only change.
.gitignore	Updates ignored entries ordering/additions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

StaticOneshotSender::send doesn’t detect the receiver-drop case: OneshotSend::send is documented to return Err(value) if the receiver was already dropped, but this implementation only errors on TrySendError::Full (which shouldn’t happen for a fresh oneshot slot). This can report Ok(()) while the value is immediately dropped on slot reclaim.

Consider checking the slot’s O_RECEIVER_ALIVE bit (and/or maintaining an explicit “receiver dropped” flag) before/after try_send and returning Err(value) when the receiver is gone, to match the transport::OneshotSend contract.

[Copilot]

StaticBoundedSender::send always returns Ok(()) and awaits Channel::send unconditionally. This violates the MpscSend contract in transport (must return Err(()) if the receiver was dropped) and can also hang indefinitely if the receiver is gone and the channel becomes full.

Consider integrating the slot’s closed flag into the send path (and waking blocked senders on receiver drop), so send() can fail fast with Err(()) when the receiver is dropped instead of always succeeding / potentially deadlocking.

[Copilot]

StaticUnboundedSender::send_now doesn’t handle the receiver-dropped case. UnboundedSend::send_now is documented to return Err(value) if the receiver was dropped, but this implementation only returns Err on TrySendError::Full.

Consider checking slot.closed (set in StaticUnboundedReceiver::drop) and returning Err(value) when closed, so callers get the same semantics as the Tokio backend and values aren’t silently accepted then dropped on reclaim.

[Copilot]

Doc comment references a crate::static_channels::static_channels! macro, but the macro introduced/used elsewhere in this PR is define_static_channels! (exported as crate::define_static_channels). This looks like a stale name and could mislead users trying to declare pools for ControlMessage.

Suggested change

	/// `crate::static_channels::static_channels!` macro names it for them.
	/// `crate::define_static_channels!` macro names it for them.

[Copilot]

The static_channels module doc mentions an “upcoming static_channels! macro”, but this PR exports define_static_channels! (via #[macro_export] in static_channels). If the macro has been renamed, updating this comment will prevent rustdoc users from searching for a non-existent macro name.

Suggested change

	/// instances that the upcoming `static_channels!` macro (phase 13.6d)
	/// instances that the `define_static_channels!` macro (phase 13.6d)

[Copilot]

This docstring mentions per-T impls “typically generated by a static_channels! macro”, but this PR exports define_static_channels! as the macro name. Updating the documentation to the correct macro name will make the compile-error guidance actionable for users.

Suggested change

	/// impls (typically generated by a `static_channels!` macro) that wire
	/// impls (typically generated by a `define_static_channels!` macro) that wire

[Copilot]

The module docs refer to a static_channels! macro (lines 11–21), but the macro provided here is define_static_channels! (see later in this file). Renaming the references would help users find the right macro and avoid confusion when following the docs.