Consider renaming `MP_STATE_PORT`

[jimmo]

Before MP_REGISTER_ROOT_POINTER was added, a port could define additional root pointers in its mpconfigport.h. It made sense to refer to them via a distinct macro, hence MP_STATE_PORT, even though all ports alias that to MP_STATE_VM.

Now the MP_REGISTER_ROOT_POINTER feature can be used to implement root pointers for modules that mpstate.h does not know about, such as modules in extmod, or user c module. We could also consider moving some config-optional stuff that's currently in mpstate.h to use MP_REGISTER_ROOT_POINTER. But things defined with MP_REGISTER_ROOT_POINTER are no longer "port". So for example, in modbluetooth_nimble.c we use MP_STATE_PORT(bluetooth_nimble_root_pointers), but the only thing that's "port" about them is that bluetooth can be enabled by the port/board.

So I see three options:

1. Remove MP_STATE_PORT and just change everything to use MP_STATE_MEM.
2. Replace MP_STATE_PORT with MP_STATE_ROOT_POINTER and have that aliased to MP_STATE_VM in a common location (i.e. not ports' mpconfigport.h).
3. Same as (2) but move the generated root pointers into their own struct, and add a fourth field (between thread and vm) on mp_state_ctx_t, and have MP_STATE_ROOT_POINTER access that specifically.

(My vote is for (3), these variables are no more "vm" than they are "port", and should find a new home )

FWIW, This came up as a question on Discord because the embed port does not define MP_STATE_PORT.

[jimmo]

cc @dlech

[dlech]

Number 3 sounds good.

I guess we don't really need MP_STATE_PORT for anything that is not a root pointer since the code can just use a local static global variable for that, so seems fine to remove it.

If we are lucky, it might even reduce code size a bit by having a smaller offset to some of the fields.

[robert-hh]

I guess we don't really need MP_STATE_PORT for anything that is not a root pointer since the code can just use a local static global variable for that, so seems fine to remove it.

None of the choices remove MP_STATE_PORT. It just get's another name and in option 3 a separate place in the root pointer data section. Besides that it's an convenient way to define a data object which can be used across different files without the need to care for a extern xyz declaration. A local static variable cannot do that.

[dlech]

None of the choices remove MP_STATE_PORT.

Sure. I meant that we don't really need it separate from root pointers. But you are right that it is that way already.

They only use case I could think of would be for really large structures with no MicroPython heap pointers that you don't want the garbage collector to traverse for performance reasons. But this is probably quite rare and as you said, we can just use extern xyz if needed for this.

[jimmo]

See #12436 which implements option (3).

@dlech sadly this did not lead to a code size decrease (the opposite it seems). So maybe I will revert the change to put it into a separate structure (but at least in the meantime that change let us find the cases where MP_STATE_VM was being used for MP_REGISTER_ROOT_POINTER fields).