[BUG] [flytekit] _ModuleSanitizer returns a wrong name when site-packages/__init__.py exists.

[amitani]

Describe the bug

When a library puts __init__.py directly under the site-packages directory (maybe accidentally), _ModuleSanitizer.get_absolute_module_name doesn't return the correct absolute module name. For example, for flytekit.core.python_auto_container.default_task_resolver, it returns site-packages.flytekit.core.python_auto_container.default_task_resolver and this causes pyflyte-execute to fail because it's passed as --resolver argument.

I encountered this when torchsurv was installed.
Not sure if this is an issue Flyte needs to solve, or this library needs to be updated. Still, I'm currently deleting site-packages/__init__.py after installing the library, and it is brittle.

A few fixes I can think of is,

1. In addition to checking if __init__.py doesn't exist in the dir, it can end the recursion when the dirname matches what's in sys.path.
2. Handle flytekit as a special case in the sanitizer.

I'm not very familiar with this part, so my suggestions may break other usage, though. Happy to contribute a PR. Some pointers are really appreciated.

(a separate issue, but a counter example in the other direction may be namespace packages. a dir without __init__.py doesn't mean it's the root)

Expected behavior

_ModuleSanitizer.get_absolute_module_name always returns flytekit.core.python_auto_container.default_task_resolver for the default task resolver, even when site-packages/__init__.py exists.

Additional context to reproduce

No response

Screenshots

No response

Are you sure this issue hasn't been raised already?

• Yes

Have you read the Code of Conduct?

• Yes