import_global does not count multiple leading dots in relative module names #73

AI-assisted source investigation, reproduced on Linux / CPython 3.12.15 against current develop at d5cfbc599fd76ed749d50fc37ba41aa5da8d2efb. This is a local reproducer, not a reported production incident.

Relative import names with two or more leading dots fail, although one-dot imports work:

import sys
import tempfile
from pathlib import Path
from python_utils.import_ import import_global

with tempfile.TemporaryDirectory() as directory:
    package = Path(directory) / 'testrel'
    (package / 'nested').mkdir(parents=True)
    (package / '__init__.py').write_text('')
    (package / 'nested' / '__init__.py').write_text('')
    (package / 'helper.py').write_text('value = 42')
    sys.path.insert(0, directory)
    try:
        scope = {'__package__': 'testrel.nested'}
        import_global('..helper', globals_=scope, locals_={})
        # ModuleNotFoundError: No module named 'testrel.nested.'
    finally:
        sys.path.remove(directory)

I expected this to import from testrel.helper and set scope['value'] to 42. The implementation removes exactly one leading empty name part and sets level = 1, leaving an extra leading dot for deeper relative names.

A private candidate counts all leading dots and removes the corresponding name parts. Its two-dot and three-dot regression rows fail on the original source; the one-dot control passes. Absolute imports and the existing explicit-level behavior are left unchanged. I checked the related #29/#46/#52/#54/#66 discussions and did not find this case covered by their compatibility decisions or the current open work.

Would a focused fix for deeper relative import names be welcome? The contribution guide recommends discussing behavior changes, so I am asking before publishing a code PR. No implementation or PR has been published, this is not an assignment claim, and no production/security impact is asserted.