* imports are not handled well

Modified patch to fix the false positive of

... may be undefined, or defined from star imports: ...

Patches should be submitted as GitHub pull requests, but a heads up that this is likely to get rejected, as imports can execute arbitrary code (and aside from that, this would slow down pyflakes by quite a bit).

Indeed. The contribution is appreciated, but pyflakes by design does not execute any of the code being checked.

I did not expect to have the patch accepted as-is. It's mainly a blue print of the needed changes, with the actual acquisition of the * module globals a proof-of-concept.

Please feel free to open a PR on GitHub, though be advised anything that imports or executes code under test will probably not be accepted. It's very slow, and can have undesirable side effects.

Even a static check would have to make some assumptions about the state of the file system and sys.path when the import happens. That could mean weird errors where pyflakes passes or fails depending on what version of a dependency happens to be installed, but a more realistic problem is that you have to deal with egg .pth files and all that nonsense.

Which isn't to say I necessarily think it can't or shouldn't be done (with some loose assumptions). But I've always felt that import * is "bad" anyway and pyflakes bailing on it isn't a huge deal.

Exception: relative import * imports in __init__.py. I really wish those worked. Maybe limiting this to only relative imports would be acceptable? I know Phil has already stated in other issues that pyflakes reading code from other files in the same module would be acceptable for pyflakes.

No worries of pyflakes failing -- if it's unable to read the * module it falls back to the default behavior.