Fix broken user types after typing-inspect 0.8.0

[vit-zikmund]

Since the recent release of typing-inspect 0.8.0, it's is_new_type() check expects the new type to come from the typing or typing_extensions module. When using local NewType, this condition isn't met.

The fix makes use of the typing.NewType and only adds the marshmallow_dataclass specific fields.

[vit-zikmund]

I made the change from github as it seemed pretty straightforward, but oh my, was I wrong. When running the local commit checks i get:

mypy.....................................................................Failed
- hook id: mypy
- exit code: 1

marshmallow_dataclass/__init__.py:809: error: Argument 1 to NewType(...) must be a string literal  [misc]
marshmallow_dataclass/__init__.py:809: error: Variable "typ" is not valid as a type  [valid-type]
marshmallow_dataclass/__init__.py:809: note: See https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases
Found 2 errors in 1 file (checked 1 source file)

Adding some ignore comments fixed the warning, but I don't have a clue if that's feasible.

[vit-zikmund]

FYI, I have no clue why the 3.8 test failed here. Can't reproduce at my place (Fedora 36) with:

python3.8 -m venv venv38
. venv38/bin/activate
pip install --upgrade pip setuptools
pip install '.[dev]'
pytest

=================================================================== test session starts ====================================================================
platform linux -- Python 3.8.13, pytest-7.1.2, pluggy-1.0.0
rootdir: /home/tlwhitec/git/github.com/lovasoa/marshmallow_dataclass, configfile: pyproject.toml
plugins: typeguard-2.13.3, mypy-plugins-1.10.0
collected 94 items                                                                                                                                         

tests/test_attribute_copy.py ....                                                                                                                    [  4%]
tests/test_city_building_examples.py ..                                                                                                              [  6%]
tests/test_class_schema.py ..............                                                                                                            [ 21%]
tests/test_collection.py ...................s                                                                                                        [ 42%]
tests/test_field_for_schema.py .......................                                                                                               [ 67%]
tests/test_forward_references.py ..........                                                                                                          [ 77%]
tests/test_mypy.yml ...                                                                                                                              [ 80%]
tests/test_optional.py ...                                                                                                                           [ 84%]
tests/test_post_load.py ..                                                                                                                           [ 86%]
tests/test_postdump.py ..                                                                                                                            [ 88%]
tests/test_union.py ...........                                                                                                                      [100%]

============================================================== 93 passed, 1 skipped in 3.29s ===============================================================

[vit-zikmund]

Investigating the error a bit I found it could be related to the way how the package is installed for the test:

pip install --pre -e '.[dev]'

This differs from the contributing guide which has only:

pip install '.[dev]'

Playing with the args, found the -e (editable) installation is a problem for mypy. Without -e, the tests end well. So maybe giving up on it for the github workflow might be a possibility? FYI @lovasoa

[lovasoa]

Why is -e a problem for mypy ? Isn't it the recommended way to install things in development mode ?

[vit-zikmund]

@lovasoa It may as well be, when you're running local tests on your workstation (so you don't have to pip install each time you change something in the code, but in case of the ephemeral nature of the github workflows, this seems to make little sense (maybe except for saving a couple ms for copying the package over to site-packages).

But the problem here is more mundane, because when there's -e, all local PR checks die on the same error (it's not just this PR). I can reproduce that locally. If I omit that -e, the tests pass. Sadly I'm not even an apprentice regarding mypy, so I don't have a clue why this breaks with -e. I came here for another reason 😏

[vit-zikmund]

Sorry, I just couldn't let it go :) It seems mypy just has some trouble figuring out the path to the package when it's installed with -e. Seems pytest doesn't use the current directory for module search by default. When I install with -e, I can still get a successful test with:

PYTHONPATH=$(pwd) pytest

or maybe even better with:

python -m pytest

which also works well. Maybe the last one is the most foolproof (it's also one of the documented ways).

[mivade]

Personally I don't think editable installs should be used on CI since that can miss things like data files not getting installed in the package (I don't believe that's an issue here though).

[vit-zikmund]

PR for fixing the mypy checks: #209 FYI @lovasoa @mivade

[dairiki]

I have cherry-picked this into #211 which has been merged to master... closing.

And thank you!