get_type_hints() fails on Mark.__init__

[Vlad-Shcherbina]

To reproduce, run the following snippet:

import typing
import _pytest.mark
print(typing.get_type_hints(_pytest.mark.Mark.__init__))

Expected:

{'return': <class 'NoneType'>, 'name': <class 'str'>, 'args': typing.List[object], 'kwargs': typing.Dict[str, object]}

Actual:

Traceback (most recent call last):
  File ".\example.py", line 3, in <module>
    print(typing.get_type_hints(_pytest.mark.Mark.__init__))
  File "C:\Python37\lib\typing.py", line 1001, in get_type_hints
    value = _eval_type(value, globalns, localns)
  File "C:\Python37\lib\typing.py", line 260, in _eval_type
    return t._evaluate(globalns, localns)
  File "C:\Python37\lib\typing.py", line 464, in _evaluate
    eval(self.__forward_code__, globalns, localns),
  File "<string>", line 1, in <module>
NameError: name 'List' is not defined

This happens because of these type annotations:

pytest/src/_pytest/mark/structures.py

Lines 122 to 129 in a48c47b

	@attr.s(frozen=True)
	class Mark(object):
	#: name of the mark
	name = attr.ib(type=str)
	#: positional arguments of the mark decorator
	args = attr.ib(type="List[object]")
	#: keyword arguments of the mark decorator
	kwargs = attr.ib(type="Dict[str, object]")

It's probably a minor issue that has no effect on the normal operation of pytest, but it could trip up runtime annotation analysis tools.

If the codebase was Python 3 only, the fix would be trivial:

from typing import List, Dict
...
@attr.s(frozen=True)
class Mark(object):
    name = attr.ib(type=str)
    args = attr.ib(type=List[object])
    kwargs = attr.ib(type=Dict[str, object])

But I'm not sure how to proceed in Python 2/3 codebase, adding the conditionals does not seem justified for such a tiny thing.

---

Version info:

• Python 3.7.0 (v3.7.0:1bf9cc5093, Jun 27 2018, 04:59:51) [MSC v.1914 64 bit (AMD64)] on win32
• pytest 3.6.2 or pytest 3.6.3.dev33+ga48c47b5

[pytestbot]

GitMate.io thinks possibly related issues are #1164 (test_something fails), #74 (test_cmdline_python_package() fails), #1852 (test_doctest.py fails ), #3048 (Pytest fails on marking of input parameter if it is a method of a class), and #2872 (mark fixtures ).

[nicoddemus]

Thanks @Vlad-Shcherbina for the report!

But I'm not sure how to proceed in Python 2/3 codebase, adding the conditionals does not seem justified for such a tiny thing.

The attr.ib docs say:

The type of the attribute. In Python 3.6 or greater, the preferred method to specify the type is using a variable annotation (see PEP 526). This argument is provided for backward compatibility. Regardless of the approach used, the type will be stored on Attribute.type.

So right now it seems the type= is only serving as the documentation, so it could be changed to:

@attr.s(frozen=True)
class Mark(object):
    name = attr.ib(type=str)
    args = attr.ib()  # type: List[object]
    kwargs = attr.ib()  # type: Dict[str, object]

@RonnyPfannschmidt what do you think?

Perhaps @hynek has another suggestion?

[hynek]

Firstly to b clear, it’s not necessarily just documentation since it’s metadata that is attached to the attributes that can be used by others – others being mostly mypy nowadays:

import attr

@attr.s
class X:
    x = attr.ib(type=int)

X("x")

leads to t.py:7: error: Argument 1 to "X" has incompatible type "str"; expected "int"

---

But it Is a non-standard way of defining types nonetheless, so I would expect type annotations to work better. Mypy knows about type= because it has an attrs plugin – CPython proper is blissfully ignorant. :)

[RonnyPfannschmidt]

typing has a backport, how about just using it?

[Vlad-Shcherbina]

If you are ok with adding extra dependency, I will prepare a PR.

[RonnyPfannschmidt]

main question is how expensive is importing it

[Vlad-Shcherbina]

It's a single 2KLOC file that imports some standard modules that are likely in the import cache already, defines a number of classes, and creates a few objects. So probably not too expensive.

from time import time
start = time()
import typing
print(time() - start)  # ~0.01s on my machine

[nicoddemus]

Hi @hynek,

Firstly to b clear, it’s not necessarily just documentation since it’s metadata that is attached to the attributes that can be used by others

Thanks for the clarification.

typing has a backport, how about just using it?

I'm a little worried about adding typing as dependency, because according to PEP-560:

The typing module is one of the heaviest and slowest modules in the standard library even with all the optimizations made. Mainly this is because subscripted generic types (see PEP 484 for definition of terms used in this PEP) are class objects (see also [1])

So much that they have made it 7 times faster in Python 3.7:

As a result of PEP 560 work, the import time of typing has been reduced by a factor of 7, and many typing operations are now faster.

I would rather use just the comments instead of adding yet another dependency which slows down pytest startup. We can always just use the typing module later when we drop 2.7 (it will be a matter of find/replace to convert from comment-based to the class-based annotations).

[hynek]

Importing typing is indeed very slow. We do all kinds of crazy shit in attrs to avoid that cost and I believe so did data classes (unless they removed it after the changes @nicoddemus mentioned).

[nicoddemus]

Thanks for the comment @hynek.

I vote to change to type comments then, at least until we drop Python 2 next year or so. @Vlad-Shcherbina what do you think? Also if you agree, would you be willing to submit a PR? 😁

[Vlad-Shcherbina]

No objections to the type comments. I'll make a PR.