Request (NOT bug): Nim Iterator export to Python Generator

[UNIcodeX]

Would it be possible to add iterator support, writing a Python "generator" as a Nim "iterator"? In my estimation, this would allow for more efficient Nim based Python libraries to be exported, rather than having to completely iterate and return a full dataset at instantiation, the iterator can lazily load when the next item is needed.

Proposal of requested functionality:

# test_lib.nim
import nimpy, tables, json

iterator myRange*(s, e: int): int {.exportpy.} =
  for num in s .. e:
    yield num

iterator loadJSON*(s: string): Table[string, string] {.exportpy.} =
  var t = initTable[string, string]()
  for i in parseJson(s):
    for key, val in i:
      t[key] = $val
    yield t

from test_lib import myRange, loadJSON

for i in myRange(1, 50):
print(i)

strJSON = """
[{
    "userId": 1,
    "name": "Ada Lovelace"
  },
  {
    "userId": 2,
    "name": "Blaise Pascal"
  }
]"""

for i in loadJSON(strJSON):
  print(i)

>>> [num for num in myRange(1, 50) if num % 2 == 0]
[2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 34, 36, 38, 40, 42, 44, 46, 48, 50]

>>> [i for i in loadJSON(strJSON)]
[{'userId': 1, 'name': 'Ada Lovelace'}, {'userId': 2, 'name': 'Blaise Pascal'}]

[UNIcodeX]

Awesome!!! Great work!!

Something I noticed in my testing on my PCPC repo is that it comes in as type class whereas a native Python generator comes in as type generator. A Nuitka generator shows type compiled_generator. This could be only due to semantics, but I'm wondering if this affects performance.

Do you know?

Implementing a generator / iterator in each of these solutions yields the following:

Running benchmark 'json_loads'.
---------------------------------------------------------
Python : 0.7985s                                    1.00x
Nuitka : 0.7754s                                    1.03x
Cython : 0.7697s                                    1.04x
Numba  : 2.9050s                                    0.27x
Nim    : 2.5040s                                    0.32x

Example with pre-built x86_64 Linux binaries (built on Arch).
json_loads.zip