module_from_spec

[mwriter]

Bug report

Bug description:

This may not be a bug, but it's a weirdness.
Why is there an automatic implicit re-registration of the builtins module in the latter case?

import sys
import importlib.util

import email

name = "email"

spec = importlib.util.find_spec(name)

sys.modules[name] = "ha"
print(sys.modules[name]) # ha

importlib.util.module_from_spec(spec)
print(sys.modules[name]) # ha

import builtins

name = "builtins"

spec = importlib.util.find_spec(name)

sys.modules[name] = "ha"
print(sys.modules[name]) # ha

importlib.util.module_from_spec(spec)
print(sys.modules[name]) # <module 'builtins' (built-in)>

CPython versions tested on:

3.12, 3.13

Operating systems tested on:

No response

[ZeroIntensity]

builtins is needed by the interpreter for all sorts of things. It probably decides to do this instead of crashing, so I don't think there's anything that can be done here.

[mwriter]

builtins is needed by the interpreter for all sorts of things. It probably decides to do this instead of crashing, so I don't think there's anything that can be done here.

This isn't true :)
I can do this

import builtins

name = "builtins"

spec = importlib.util.find_spec(name)

sys.modules[name] = "ha"
print(sys.modules[name]) # ha

importlib.util.module_from_spec(spec)
print(sys.modules[name]) # ha

Perhaps I will present this part of my research soon. Although it was bullied on the main forum.

[ZeroIntensity]

Yes, because you don't overwrite builtins there. The interpreter intentionally keeps its own copy of builtins, it doesn't surprise me that it swaps it out if it needs to.

[ZeroIntensity]

In general, black magic with sys.modules won't be encouraged or fixed on the Python side. I suggest for both of your issues you use the private C API to manually mess with things, you have more power there.

[mwriter]

The interpreter intentionally keeps its own copy of builtins

According to my research, all copies can be overwritten.

In general, black magic with sys.modules won't be encouraged or fixed on the Python side

There are no problems for me here. At the python level, I can change everything myself.
But there is a clear discrepancy in the logic of the code's behavior.

I suggest for both of your issues you use the private C API to manually mess with things

For example?

[ZeroIntensity]

According to my research, all copies can be overwritten.

Sure, but that doesn't mean we have to support it :)

For example?

Truly overwriting the sys.modules:

#define Py_BUILD_CORE
#include "pycore_interp.h"

static PyObject *
overwrite_sys_modules(PyObject *self, PyObject *obj)
{
    PyInterpreterState *interp = PyInterpreterState_Get();
    Py_SETREF(interp->imports.modules, Py_NewRef(obj));
    Py_RETURN_NONE;
}

[mwriter]

Sure, but that doesn't mean we have to support it

This is understandable. It was only in the context of the behavior inconsistency in the original example here. But for me it's not a problem at all. I just decided to write :)

Truly overwriting the

Thanks for the example.