Stubtest whines about Concatenate given as Callable argument

[JulienPalard]

With the concatenate example from the typing doc, which works with mypy 0.950, if I split it to a file and a stub, stubtest starts to compain about Concatenate being used as a Callable argument:

totest.py:

from threading import Lock

# Use this lock to ensure that only one thread is executing a function
# at any time.
my_lock = Lock()

def with_lock(f):
    '''A decorator which provides a lock.'''
    def inner(*args, **kwargs):
        # Provide the lock as the first argument.
        return f(my_lock, *args, **kwargs)
    return inner

@with_lock
def sum_threadsafe(lock, numbers):
    '''Add a list of numbers together in a thread-safe manner.'''
    with lock:
        return sum(numbers)

# We don't need to pass in the lock ourselves thanks to the decorator.
sum_threadsafe([1.1, 2.2, 3.3])

totest.pyi:

from threading import Lock
from typing import Concatenate, ParamSpec, TypeVar, Callable

P = ParamSpec('P')
R = TypeVar('R')


def with_lock(f: Callable[Concatenate[Lock, P], R]) -> Callable[P, R]: ...

stubtest run:

$ stubtest totest
error: not checking stubs due to mypy build errors:
totest.pyi:8: error: The first argument to Callable must be a list of types or "..."

[hauntsaninja]

Thanks for the report, but I can't repro. Are you sure you're using mypy 0.950? Consider using python -m mypy.stubtest totest as the invocation.

I do get:

error: totest.P is an unknown mypy node
Stub: at line 5
ParamSpecExpr:5()
Runtime:
MISSING

error: totest.my_lock is not present in stub
Stub:
MISSING
Runtime:
<unlocked _thread.lock object at 0x1124fa090>

error: totest.sum_threadsafe is not present in stub
Stub:
MISSING
Runtime: at line 9 in file /Users/shantanu/devoai/openai/totest.py
<function with_lock.<locals>.inner at 0x10e94ce50>

where the first error is something stubtest could improve.

The error you're getting is actually not stubtest specific, it's from the underlying mypy build. mypy itself only gained support for Concatenate in 0.950.

[JulienPalard]

Thanks for trying my reproducer @hauntsaninja. As you, I can't repro today. What I think happened is that I had a mypy 0.950 in the venv (clearly remember mypy --version telling me so) but for reasons bash's hash got me another stubtest (I also have a user install of mypy). TIL about python -m mypy.stubtest, it would have avoided me and you the issue. Sorry for the noise.

Maybe adding a --version to stubtest would be helpfull for those cases.

[hauntsaninja]

That's a good suggestion! And thanks for reporting, despite the lack of repro, a number of things have come from this:

• Better error from mypy about Callable types that make it clearer that ParamSpec should work Friendlier errors for PEP 612 #12832
• Adding --version to stubtest stubtest: add --version #12852
• Fixing the "unknown mypy node" that I can repro above stubtest: check typevar and paramspec #12851