# \[WASI\] Importing namespaced 3rd party C extensions (with import cycles)

**URL:** <https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606>\
**Category:** WebAssembly\
**Created:** [March 8, 2023, 4:08pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606 "2023-03-08T16:08:07Z")\
**Posts on this page:** 20\
**Page:** 1

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 8, 2023, 4:08pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/1 "2023-03-08T16:08:07Z")

</div>

I’ve been trying to get C extensions working in the WASI build of Python. I started with a simple one that lives at the top-level, statically linked it, and used PyImport\_AppendInittab in my C wrapper and it worked just fine. I’ve moved on to a more difficult test: numpy. I’ve been able to get numpy to compile to Wasm as static archives and I can link them with a static Python library, however, I’m having a problem importing numpy.core.\_multiarray\_umath.

Importing that library seems to cause a cyclical import. Its module initialization function imports numpy.core.\_exceptions which causes some other imports to happen, eventually reaching numpy.core.multiarray which imports numpy.core.\_multiarray\_umath again. In the standard Python build, this works fine and it just continues, but in the Wasi build it fails when it hits that import cycle. I’m not clear as to what happens that the standard build works that isn’t happening in the Wasi version (it likely has to do with static linking, not so much Wasi). If anyone has any tips, I’d greatly appreciate it.

---

<div class="post-metadata">

**Author:** ![brettcannon](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/brettcannon/32/34895_2.png) [@brettcannon](https://discuss.python.org/u/brettcannon)\
**Post date:** [March 8, 2023, 8:45pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/2 "2023-03-08T20:45:43Z")

</div>

> [@kesmit](#):
>
> I’ve been trying to get C extensions working in the WASI build of Python. I started with a simple one that lives at the top-level, statically linked it, and used PyImport\_AppendInittab in my C wrapper and it worked just fine.

That’s great to hear! I have been thinking that might be the best solution for WASI until we get some dynamic linking solution (I’m hoping the component model could get leveraged for this some day).

> [@kesmit](#):
>
> Importing that library seems to cause a cyclical import. Its module initialization function imports numpy.core.\_exceptions which causes some other imports to happen, eventually reaching numpy.core.multiarray which imports numpy.core.\_multiarray\_umath again. In the standard Python build, this works fine and it just continues, but in the Wasi build it fails when it hits that import cycle. I’m not clear as to what happens that the standard build works that isn’t happening in the Wasi version (it likely has to do with static linking, not so much Wasi).

The code that handles built-in imports is in the [`BuiltinImporter` class](https://github.com/python/cpython/blob/cbb0aa71d040022db61390380b8aebc7c04f3275/Lib/importlib/_bootstrap.py#L877). There’s nothing specifically there that I can see which would suggest statically linking and using built-in imports would cause a failure that would work otherwise. Typically cyclical imports fail because someone is importing to a specific object in a module instead of the module itself since the import system sets a module in `sys.modules` ASAP specifically so any other imports will get the module from there and not trigger another import.

Looking at the [`create_builtin()` function](https://github.com/python/cpython/blob/cbb0aa71d040022db61390380b8aebc7c04f3275/Python/import.c#L1173), it _might_ be the cause **if** numpy is using single-phase initialization. If I’m tracing through the code by hand correctly, I think [cpython/Python/import.c at cbb0aa71d040022db61390380b8aebc7c04f3275 · python/cpython · GitHub](https://github.com/python/cpython/blob/cbb0aa71d040022db61390380b8aebc7c04f3275/Python/import.c#L1044) will never succeed for anything that’s being actively imported in NumPy because [cpython/Python/import.c at cbb0aa71d040022db61390380b8aebc7c04f3275 · python/cpython · GitHub](https://github.com/python/cpython/blob/cbb0aa71d040022db61390380b8aebc7c04f3275/Python/import.c#L1204) is where a module gets inserted into that specific single-phase module cache in `create_builtin()`. And since that’s passed [where the module’s init function gets called](https://github.com/python/cpython/blob/cbb0aa71d040022db61390380b8aebc7c04f3275/Python/import.c#L1188), it’s going to lead to a stack of calls into `create_builtin()` where no one has gotten into a place to break any cycle.

Any chance you can take your build and run it under a debugger to see if my hunch is correct? If it is then please open an issue at [Issues · python/cpython · GitHub](https://github.com/python/cpython/issues) ? I would also then open an issue on NumPy letting them know.

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 8, 2023, 10:15pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/3 "2023-03-08T22:15:13Z")

</div>

I haven’t had much luck getting a debugger to work, so I’ve had to track things down the old fashioned way (printf…). The line where things go bad is [https://github.com/python/cpython/blob/main/Python/import.c#L2547](https://github.com/python/cpython/blob/main/Python/import.c#L2547). The first time this is hit by importing numpy.core.\_multiarray\_umath, it causes the parents to get imported (numpy.core and numpy) which causes the subsequent imports to come around to importing numpy.core.\_multiarray\_umath again. The second time it hits that line in import.c, it fails.

I’ll keep trying to get a debugger on it, and maybe post a repo with my changes and build scripts so that you can try to reproduce it if you have time.

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 8, 2023, 10:17pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/4 "2023-03-08T22:17:02Z")

</div>

As far as the working C extension goes, it came together relatively easily. I even have exports from the host program that are called by the Python C extension, so it’s demonstrating communication both ways from the host.

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 9, 2023, 3:45pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/5 "2023-03-09T15:45:29Z")

</div>

Unfortunately, it looks like debugging may not be an option.

[https://github.com/bytecodealliance/wasmtime/issues/4669](https://github.com/bytecodealliance/wasmtime/issues/4669)

---

<div class="post-metadata">

**Author:** ![brettcannon](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/brettcannon/32/34895_2.png) [@brettcannon](https://discuss.python.org/u/brettcannon)\
**Post date:** [March 9, 2023, 10:44pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/6 "2023-03-09T22:44:41Z")

</div>

> [@kesmit](#):
>
> I haven’t had much luck getting a debugger to work, so I’ve had to track things down the old fashioned way (printf…).

Have you tried doing the same build to the native CPU of your machine so you can debug that way? In other words don’t build for WASI for the target but your machine’s own CPU, leaving everything else the same?

> [@kesmit](#):
>
> The line where things go bad is [https://github.com/python/cpython/blob/main/Python/import.c#L2547](https://github.com/python/cpython/blob/main/Python/import.c#L2547).

Can you provide a permalink? That line currently goes to a `}` in the middle of a function.

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 9, 2023, 11:04pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/7 "2023-03-09T23:04:07Z")

</div>

Sorry, here’s the link: [https://github.com/python/cpython/blob/c6858d1e7f4cd3184d5ddea4025ad5dfc7596546/Python/import.c#L2552](https://github.com/python/cpython/blob/c6858d1e7f4cd3184d5ddea4025ad5dfc7596546/Python/import.c#L2552).

I’ve been trying to build a native version of Python and numpy, but I’m having trouble getting it to work for some reason that I haven’t figured out yet.

---

<div class="post-metadata">

**Author:** ![assambar](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/assambar/32/10926_2.png) [@assambar](https://discuss.python.org/u/assambar)\
**Post date:** [March 10, 2023, 2:12pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/8 "2023-03-10T14:12:08Z")

</div>

> [@kesmit](#):
>
> Unfortunately, it looks like debugging may not be an option.
> 
> [https://github.com/bytecodealliance/wasmtime/issues/4669](https://github.com/bytecodealliance/wasmtime/issues/4669)

@kesmit I was hitting this last year and I just removed the assert and changed the if above it as in

```auto
- if range_start == range_end {
+ if range_start >= range_end {
                 continue;
             }
- assert!(range_start < range_end);

```

Then, with a local build of wasmtime I was able to do some debugging. It’s not as good as printf, but did give me faster iteration cycles till I find the places worthy of a printf.

I have no idea what those ranges represent, just saying that it worked for me.

---

<div class="post-metadata">

**Author:** ![brettcannon](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/brettcannon/32/34895_2.png) [@brettcannon](https://discuss.python.org/u/brettcannon)\
**Post date:** [March 11, 2023, 12:15am UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/9 "2023-03-11T00:15:55Z")

</div>

> [@kesmit](#):
>
> Sorry, here’s the link: [cpython/Python/import.c at c6858d1e7f4cd3184d5ddea4025ad5dfc7596546 · python/cpython · GitHub](https://github.com/python/cpython/blob/c6858d1e7f4cd3184d5ddea4025ad5dfc7596546/Python/import.c#L2552)

That’s calling [cpython/Lib/importlib/\_bootstrap.py at 90f1d777177e28b6c7b8d9ba751550e373d61b0a · python/cpython · GitHub](https://github.com/python/cpython/blob/90f1d777177e28b6c7b8d9ba751550e373d61b0a/Lib/importlib/_bootstrap.py#L1296) .

What’s the exact failure you’re seeing? Are you blowing out your stack due to infinite recursion?

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 13, 2023, 6:17pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/10 "2023-03-13T18:17:48Z")

</div>

I did get a little further by getting a static x86 build working. I found out that there was a linker option that needed to be added. That got the x86 build working, but unfortunately, it didn’t completely fix the Wasm build (although it is getting a little further). I’m still digging…

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 13, 2023, 9:10pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/11 "2023-03-13T21:10:07Z")

</div>

@brettcannon I’m struggling to debug the BuiltinImporter code because it is frozen. Is there a way to compile without freezing it?

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 13, 2023, 9:22pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/12 "2023-03-13T21:22:08Z")

</div>

I thought I would at least be able to edit Lib/importlib/\_bootstrap.py with print statements, rebuild Python and they would show up, but that’s not even working so there’s something I don’t understand here.

---

<div class="post-metadata">

**Author:** ![brettcannon](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/brettcannon/32/34895_2.png) [@brettcannon](https://discuss.python.org/u/brettcannon)\
**Post date:** [March 13, 2023, 11:12pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/13 "2023-03-13T23:12:50Z")

</div>

There’s a regen-all make target to regenerate the frozen copy of `importlib`. You can also use `importlib.import_module()` and that uses the `.py` files.

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 14, 2023, 7:32pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/14 "2023-03-14T19:32:33Z")

</div>

```auto
import wasmtime
from bindings import Udf

store = wasmtime.Store()

module = wasmtime.Module(store.engine, open('s2-udf-python3.10.wasm', 'rb').read())

linker = wasmtime.Linker(store.engine)
linker.define_wasi()

wasi = wasmtime.WasiConfig()
wasi.inherit_stdin()
wasi.inherit_stdout()
wasi.inherit_stderr()
wasi.env = [
    ('PYTHONDONTWRITEBYTECODE', 'x'),
# ('PYTHONVERBOSE', '5'),
]
store.set_wasi(wasi)

udf = Udf(store, linker, module)

# You *must* call _initialize for wasi to work
udf.instance.exports(store)['_initialize'](store)

udf.exec(store, r'''
import platform
import numpy as np

print('platform', platform.platform())
print('numpy', np. __version__ )
print('')

print('a')
a = np.array([2, 3, 4])
print(a)
print(a.dtype)
print('')

print('b')
b = np.array([1.2, 3.5, 5.1])
print(b)
print(b.dtype)
print('')

print('zeros')
print(np.zeros((3, 4)))
print('')

print('ones')
print(np.ones((2, 3, 4)))
print('')

print('empty')
print(np.empty((2, 3)))
print('')

print('int range')
print(np.arange(10, 30, 5))
print('')

print('float range')
print(np.arange(0, 2, 0.3))
print('')

print('linspace')
print(np.linspace(0, 2, 9))
x = np.linspace(0, 2 * np.pi, 100)
f = np.sin(x)
print(f)
print('')

print('a - b')
print(a - b)
print('')

print('b**2')
print(b ** 2)
print('')
''')

```

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 14, 2023, 7:33pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/15 "2023-03-14T19:33:00Z")

</div>

```auto
platform wasi-0.0.0-wasm32-32bit
numpy 1.24.2

a
[2 3 4]
int32

b
[1.2 3.5 5.1]
float64

zeros
[[0. 0. 0. 0.]
 [0. 0. 0. 0.]
 [0. 0. 0. 0.]]

ones
[[[1. 1. 1. 1.]
  [1. 1. 1. 1.]
  [1. 1. 1. 1.]]

 [[1. 1. 1. 1.]
  [1. 1. 1. 1.]
  [1. 1. 1. 1.]]]

empty
[[3.24640964e-283 7.63566008e-227 2.59665904e-225]
 [9.70076201e-225 1.46511924e-225 9.70096813e-225]]

int range
[10 15 20 25]

float range
[0. 0.3 0.6 0.9 1.2 1.5 1.8]

linspace
[0. 0.25 0.5 0.75 1. 1.25 1.5 1.75 2.]
[ 0.00000000e+00 6.34239197e-02 1.26592454e-01 1.89251244e-01
  2.51147987e-01 3.12033446e-01 3.71662456e-01 4.29794912e-01
  4.86196736e-01 5.40640817e-01 5.92907929e-01 6.42787610e-01
  6.90079011e-01 7.34591709e-01 7.76146464e-01 8.14575952e-01
  8.49725430e-01 8.81453363e-01 9.09631995e-01 9.34147860e-01
  9.54902241e-01 9.71811568e-01 9.84807753e-01 9.93838464e-01
  9.98867339e-01 9.99874128e-01 9.96854776e-01 9.89821442e-01
  9.78802446e-01 9.63842159e-01 9.45000819e-01 9.22354294e-01
  8.95993774e-01 8.66025404e-01 8.32569855e-01 7.95761841e-01
  7.55749574e-01 7.12694171e-01 6.66769001e-01 6.18158986e-01
  5.67059864e-01 5.13677392e-01 4.58226522e-01 4.00930535e-01
  3.42020143e-01 2.81732557e-01 2.20310533e-01 1.58001396e-01
  9.50560433e-02 3.17279335e-02 -3.17279335e-02 -9.50560433e-02
 -1.58001396e-01 -2.20310533e-01 -2.81732557e-01 -3.42020143e-01
 -4.00930535e-01 -4.58226522e-01 -5.13677392e-01 -5.67059864e-01
 -6.18158986e-01 -6.66769001e-01 -7.12694171e-01 -7.55749574e-01
 -7.95761841e-01 -8.32569855e-01 -8.66025404e-01 -8.95993774e-01
 -9.22354294e-01 -9.45000819e-01 -9.63842159e-01 -9.78802446e-01
 -9.89821442e-01 -9.96854776e-01 -9.99874128e-01 -9.98867339e-01
 -9.93838464e-01 -9.84807753e-01 -9.71811568e-01 -9.54902241e-01
 -9.34147860e-01 -9.09631995e-01 -8.81453363e-01 -8.49725430e-01
 -8.14575952e-01 -7.76146464e-01 -7.34591709e-01 -6.90079011e-01
 -6.42787610e-01 -5.92907929e-01 -5.40640817e-01 -4.86196736e-01
 -4.29794912e-01 -3.71662456e-01 -3.12033446e-01 -2.51147987e-01
 -1.89251244e-01 -1.26592454e-01 -6.34239197e-02 -2.44929360e-16]

a - b
[0.8 -0.5 -1.1]

b**2
[1.44 12.25 26.01]

```

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 14, 2023, 7:37pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/16 "2023-03-14T19:37:36Z")

</div>

@brettcannon There was an issue in the Python BuiltinImporter. In this case, the builtin extension was namespaced below `numpy.core`. The BuiltinImport is not expecting namespaced extensions and when it sees the `path` parameter supplied, it short-circuits and returns `None`. The simple fix I put in (that may not be perfect), is to check for a ‘.’ in the import name and to see if it is a builtin name before doing anything else.

```auto
class BuiltinImporter:

   . . .

    @classmethod
    def find_spec(cls, fullname, path=None, target=None):
        if '.' in fullname and _imp.is_builtin(fullname):
            return spec_from_loader(fullname, cls, origin=cls._ORIGIN)
        if path is not None:
            return None
        if _imp.is_builtin(fullname):
            return spec_from_loader(fullname, cls, origin=cls._ORIGIN)
        else:
            return None

```

---

<div class="post-metadata">

**Author:** ![brettcannon](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/brettcannon/32/34895_2.png) [@brettcannon](https://discuss.python.org/u/brettcannon)\
**Post date:** [March 14, 2023, 8:05pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/17 "2023-03-14T20:05:51Z")

</div>

> [@kesmit](#):
>
> There was an issue in the Python BuiltinImporter. In this case, the builtin extension was namespaced below `numpy.core`. The BuiltinImport is not expecting namespaced extensions and when it sees the `path` parameter supplied, it short-circuits and returns `None`. The simple fix I put in (that may not be perfect), is to check for a ‘.’ in the import name and to see if it is a builtin name before doing anything else.

Yeah, that’s a mistake. 😅 Would you mind opening an issue at [Issues · python/cpython · GitHub](https://github.com/python/cpython/issues) ? I can then fix that upstream.

---

<div class="post-metadata">

**Author:** ![kesmit](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/kesmit/32/11083_2.png) [@kesmit](https://discuss.python.org/u/kesmit)\
**Post date:** [March 16, 2023, 3:21pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/18 "2023-03-16T15:21:48Z")

</div>

Here’s the next step: pandas. However, pandas has some much bigger problems than numpy. It only sorta works, and the build takes more hand-holding.

```auto
import platform
print(platform.platform())

import pandas as pd
df = pd.DataFrame([[1.234, 2, 'hi'], [3.14, 3, 'bye']], columns=['a', 'b', 'c'])
print('COLUMNS', df.columns)
#print('DTYPES', df.dtypes)
for i, row in enumerate(df.iterrows()):
    print('ROW', i, row[1])

```

Result:

```auto
wasi-0.0.0-wasm32-32bit
COLUMNS Index(['a', 'b', 'c'], dtype='object')
ROW 0 a 1.234
b 2
c hi
Name: 0, dtype: object
ROW 1 a 3.14
b 3
c bye
Name: 1, dtype: object

```

---

<div class="post-metadata">

**Author:** ![brettcannon](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/brettcannon/32/34895_2.png) [@brettcannon](https://discuss.python.org/u/brettcannon)\
**Post date:** [March 16, 2023, 9:45pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/19 "2023-03-16T21:45:47Z")

</div>

> [@kesmit](#):
>
> Here’s the next step: pandas.

Out of curiosity, are you adding pandas due to demand or because you just want to? I totally get NumPy as it’s so foundational for the sciences/numerics, but I would interested if pandas is there or if pure Python alternatives could also work in this instance to save the headache of trying to make it work.

---

<div class="post-metadata">

**Author:** ![brettcannon](https://sea2.discourse-cdn.com/flex002/user_avatar/discuss.python.org/brettcannon/32/34895_2.png) [@brettcannon](https://discuss.python.org/u/brettcannon)\
**Post date:** [March 16, 2023, 9:48pm UTC](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606/20 "2023-03-16T21:48:31Z")

</div>

> [@brettcannon](#):
>
> Would you mind opening an issue at [Issues · python/cpython · GitHub](https://github.com/python/cpython/issues) ? I can then fix that upstream.

I didn’t want to wait on this, so I opened [Allow for built-in modules to be submodules · Issue #102768 · python/cpython · GitHub](https://github.com/python/cpython/issues/102768) .

[Next page](https://discuss.python.org/t/wasi-importing-namespaced-3rd-party-c-extensions-with-import-cycles/24606.md?page=2)