This is apparently good change, especially if it leads to code savings. Would need to review in more detail regarding what's included and what's not.

Some things to think about alongside this change:

1. This change apparently mandates architecture that each VM would have own heap. Maybe worth supporting shared heap, i.e. have a separate structure for heap variables?
2. Can we also think about optimzing gc roots scanning? Current scanning of entire .bss is sub-ideal - for example, having a static heap array puts it into .bss and then there's no real gc.

Glad you agree this is a good path to follow. The PR is messy, but the idea is there.

This change apparently mandates architecture that each VM would have own heap. Maybe worth supporting shared heap, i.e. have a separate structure for heap variables?

Yes, we can have a (sub)struct for heap/gc variables.

Can we also think about optimzing gc roots scanning? Current scanning of entire .bss is sub-ideal - for example, having a static heap array puts it into .bss and then there's no real gc.

Yes, definitely. It was on my agenda to have different structs for data, bss, and bss-containing-root-pointers. Just didn't know the cleanest way to do it yet.

It also includes a partial move away from Plan 9 style header includes

That might become a messy git log afterwards which becomes harder to investigate due to mixing (sort of) seperate issues. If the plan is to move away from Plan 9 anyway, consider doing that in one single commit and then build upon that?

For the rest: there should be only benefits to reducing the amount of global state so this is definitely a good thing

If the plan is to move away from Plan 9 anyway, consider doing that in one single commit and then build upon that?

Yeah, ok. I'll do a separate PR for that.

Note that, to support proper globals/locals context switching when calling functions, we should have an "outer" global state object which is small, easy to fit on the stack, and easy to change globals/locals. Something like:

struct mpy_ctx_t {
    mp_obj_dict_t *globals, *locals;
    mp_heap_state_t *heap_state;
    mp_interp_state_t *interp_state;
};

Possibly don't need to separate heap_state and interp_state here.

This PR is now waiting on #1022.

Superseded by #1032.