Hide imported names in stubs unless 'as id' is used (PEP 484)

[ilevkivskyi]

Fixes #2927

PEP 484 specifies that only names imported as from mod import name as name should be re-exported. However, mypy didn't follow this rule so that this code passed without errors but obviously fails at runtime:

from collections import TypeVar
from weakref import Generic
from getopt import List

T = TypeVar('T')

class C(Generic[T], List[int]):
    ...

This PR makes all these errors. CI will fail since there are several dozens of errors in typeshed, I am going to make a PR to typeshed soon.

[gvanrossum]

I'll test this against Dropbox codebases.

[ilevkivskyi]

It looks like some of the typeshed failures are due to a bug in my PR, will investigate now.

[gvanrossum]

Yeah, I get this for example:

mypy -2 -c 'import xml.sax'
typeshed/stdlib/2and3/xml/sax/__init__.pyi:27: error: Name 'xml.sax.xmlreader.XMLReader' is not defined
typeshed/stdlib/2and3/xml/sax/__init__.pyi:29: error: Name 'xml.sax.handler.ContentHandler' is not defined
typeshed/stdlib/2and3/xml/sax/__init__.pyi:30: error: Name 'xml.sax.handler.ErrorHandler' is not defined
typeshed/stdlib/2and3/xml/sax/__init__.pyi:32: error: Name 'xml.sax.handler.ContentHandler' is not defined
typeshed/stdlib/2and3/xml/sax/__init__.pyi:33: error: Name 'xml.sax.handler.ErrorHandler' is not defined
typeshed/stdlib/2and3/xml/sax/__init__.pyi:35: error: Name 'xml.sax.xmlreader.XMLReader' is not defined

[ilevkivskyi]

After I fixed the bug in this PR, it looks like there are only two errors in typeshed:

error: Module 'json' has no attribute 'JSONDecodeError'
error: Module 'asyncio.futures' has no attribute 'Awaitable'

Both seem to be legitimate errors.

[ilevkivskyi]

only two errors in typeshed

Actually two in stdlib plus a dozen or so in third party stubs.

[gvanrossum]

So do we need to fix all the typeshed errors first? Also I can't re-test this with the Dropbox codebases until you've fixed the merge conflict.

[ilevkivskyi]

@gvanrossum I fixed the merge conflict and made a PR to typeshed python/typeshed#1484

[gvanrossum]

Please hold off on this until after the release of 0.521 (or until Jukka vets it).

[ilevkivskyi]

Please hold off on this until after the release of 0.521 (or until Jukka vets it).

NP, I don't think this is urgent. But it would be great to test this with internal codebases (maybe there are more errors in stubs).

[ilevkivskyi]

OK, I have fixed the logic as discussed in python/typeshed#1484, this should be now ready for review and testing with internal codebases.

[ilevkivskyi]

I added one more test and simplified the PR, still I cannot do this with only one flag module_public since it is used for other things like __all__, import * and _private_names. It is possible to have a single flag with three values HIDDEN, PRIVATE, PUBLIC = 0, 1, 2 instead of two bool flags module_public and module_hidden but this will induce some code churn.

Also typeshed is synced, so that this ready to be reviewed and merged.

[emmatyping]

One style nit, otherwise looks good to me.

[emmatyping]

Minor style nit. Take it if you want. I prefer to not have compound if statements in returns. It makes things clearer if you break it up, but up to you.

[emmatyping]

Thanks! Looks good.

[ilevkivskyi]

@gvanrossum Do you think Ethan's review is enough, or we need another one? (This is quite a minor change, so that I would merge this soon to avoid more merge conflicts.)

[gvanrossum]

I'm taking a look now.

[gvanrossum]

LGTM. I'm pressed for time, can you merge it yourself?