Does Cython do stack allocation of final extension types? Does it do allocation inline of these types in other extension types? What about C structs?

Consider following Cython code:

import cython

@cython.final
cdef class Fish:
   cdef double weight

cdef class Aquarium:
   cdef Fish fish

cpdef double fish_weight(Fish fish):
   return fish.weight

cpdef void change_fish_weight(Fish fish, double weight):
   fish.weight = weight

cpdef double new(double weight):
   cdef Fish f = Fish()
   f.weight = weight
   return f.weight


cdef Aquarium aq = Aquarium()
cdef Fish fish = Fish()

change_fish_weight(fish, 5.0)
fish_weight(fish)
new(5.0)

1. Would be Fish allocated inline in aq?
2. Would be fish "passed by reference" to change_fish_weight?
3. Would be fish "passed by reference" to fish_weight?
4. Would be f allocated on stack in new function? (Let's suppose optimizer won't entirely eliminate variable f as needless.)
5. What about previous question if Fish was struct and not extension type?

Which answers change, if compiler optimization is enabled? Why?

Would be Fish allocated inline in aq?

No

Would be fish "passed by reference" to change_fish_weight?

Yes

Would be fish "passed by reference" to fish_weight?

Yes

Would be f allocated on stack in new function? (Let's suppose optimizer won't entirely eliminate variable f as needless.)

No.

What about previous question if Fish was struct and not extension type?

A struct would be allocated on the stack.

Which answers change, if compiler optimization is enabled? Why?

None of the answers change.

--------------------------------------------

Essentially final extension types are still Python objects and so are still heap allocated because they have to be passed around by reference like any other Python object. This answer explains the reasons to use final extension types - they're relatively small optimizations and have nothing to do with stack allocation.