pylint-dev / pylint

It's not just a linter that annoys you!
https://pylint.readthedocs.io/en/latest/
GNU General Public License v2.0
5.33k stars 1.14k forks source link

Pylint can't find some PyGObject classes' members #6352

Open Scrashdown opened 2 years ago

Scrashdown commented 2 years ago

Bug description

Hello,

I have a simple class inheriting from PyGObject's Gtk.Window class. Pylint cannot Gtk.Window's add and show_all members, yet the code runs fine:

"""
The main window of the application.
"""

from typing import Tuple

import gi
gi.require_version("Gtk", "3.0")
# pylint: disable=wrong-import-position
from gi.repository import Gtk
# pylint: enable=wrong-import-position

class MainWindow(Gtk.Window):
    """
    The main window of the application.
    """

    WINDOW_SIZE: Tuple[int, int] = (500, 250)
    MAIN_PADDING_PX: int = 10
    FRAMES_VERTICAL_SPACE_PX: int = 10

    INPUT_FILE_FRAME_LABEL: str = "Input file"
    INPUT_FILE_BUTTON_LABEL: str = "Choose input file"

    def __init__(self, title: str):
        super().__init__(title=title)
        self.set_size_request(*self.WINDOW_SIZE)

        # --- Add widgets --- #

        # Main container
        main_box: Gtk.Box = Gtk.Box.new(orientation=Gtk.Orientation.VERTICAL,
                                        spacing=self.FRAMES_VERTICAL_SPACE_PX)
        self.add(main_box)
        MainWindow.__pad_on_all_sides(main_box, self.MAIN_PADDING_PX)

        # Input file picker
        input_file_frame: Gtk.Frame = Gtk.Frame.new(label=self.INPUT_FILE_FRAME_LABEL)
        main_box.pack_start(child=input_file_frame, expand=True, fill=True, padding=0)

        hbox: Gtk.Box = Gtk.Box.new(orientation=Gtk.Orientation.HORIZONTAL)
        input_file_frame.add(hbox)

        self.input_file_button = Gtk.Button.new(label=self.INPUT_FILE_BUTTON_LABEL)

        # --- Connect and show --- #
        self.connect("destroy", Gtk.main_quit)
        self.show_all()

    @staticmethod
    def __pad_on_all_sides(widget: Gtk.Widget, padding: int) -> None:
        widget.set_margin_bottom(padding)
        widget.set_margin_top(padding)
        widget.set_margin_left(padding)
        widget.set_margin_right(padding)

Interestingly, if I remove the following line, pylint stops complaining. It looks like adding a member to the MainWindow class which inherits from Gtk.Window throws off pylint.

self.input_file_button = Gtk.Button.new(label=self.INPUT_FILE_BUTTON_LABEL)

Configuration

I am using pylint's default configuration.

Command used

pylint main_window.py

Pylint output

************* Module main_window
main_window.py:34:8: E1101: Instance of 'MainWindow' has no 'add' member (no-member)
main_window.py:48:8: E1101: Instance of 'MainWindow' has no 'show_all' member (no-member)

------------------------------------------------------------------
Your code has been rated at 6.43/10 (previous run: 6.43/10, +0.00)

Expected behavior

I would expect pylint to find these two members, as they clearly exist since the code runs fine.

Pylint version

pylint 2.13.5
astroid 2.11.2
Python 3.10.4 (main, Mar 23 2022, 23:05:40) [GCC 11.2.0]

OS / Environment

EndeavourOS KDE version 5.17.3-arch1-1 kernel All packages up-to-date

Additional dependencies

astroid==2.11.2 black==22.3.0 click==8.1.2 dill==0.3.4 isort==5.10.1 lazy-object-proxy==1.7.1 mccabe==0.7.0 mypy==0.942 mypy-extensions==0.4.3 pathspec==0.9.0 platformdirs==2.5.1 pycairo==1.21.0 PyGObject==3.42.0 pylint==2.13.5 tomli==2.0.1 typing_extensions==4.1.1 wrapt==1.14.0

jacobtylerwalls commented 2 years ago

Hi @Scrashdown, thanks for the report.

The astroid project handles most of the static analysis hints for specific projects to help pylint guess about membership. Unfortunately the "brain" for gi is completely untested. If you (or anyone else!) have got a cycle for it, there's a model for how to get a new package-specific unittest written in PyCQA/astroid#1505.

hadess commented 2 years ago

The astroid project handles most of the static analysis hints for specific projects to help pylint guess about membership. Unfortunately the "brain" for gi is completely untested. If you (or anyone else!) have got a cycle for it, there's a model for how to get a new package-specific unittest written in PyCQA/astroid#1505.

Based on that, this seems to work correctly with python3-astroid-2.11.6-1.fc36.noarch but fails with python3-astroid-2.12.1-1.fc37, so it would be a regression.

hadess commented 1 year ago

Seems to work correctly with pylint-2.14.4-3.fc37.noarch and python3-astroid-2.12.13-1.fc37.noarch for me.

DanielNoord commented 1 year ago

So, this is then no longer an issue?

hadess commented 1 year ago

Not for me, but I'm not the original reporter.

Salamandar commented 1 year ago

I can confirm this issue is still present with those versions :

pylint 2.17.0
astroid 2.15.0
Python 3.10.9 (main, Dec 19 2022, 17:35:49) [GCC 12.2.0]

I have the same issue on my code AND on the original reporter's code.

kulikjak commented 1 year ago

Hi, I've also hit this issue. The following code is the shortest reproducible case I found.

import gi
gi.require_version('Gtk', '3.0')
from gi.repository import Gtk

cell = Gtk.CellRendererText()
#cell.props.xalign = 1.0

Gtk.Builder().connect_signals({})

When you run it like this, it works as expected; however, when you uncomment the cell.props.xalign line, it breaks with test.py:8:0: E1101: Instance of 'Builder' has no 'connect_signals' member (no-member) (connect_signals is only available in Gtk3, not in Gtk4)

I did a little bit more digging and found out that in the second (failing) case, following exception is raised in _register_require_version: Namespace Gtk is already loaded with version 4.0

Debug prints in each function in brain_gi.py further showed that in the first case, require_version is handled first as expected. In the second case, _register_require_version is called after everything else has been processed, which is wrong (newer Gtk was already loaded at this point).

I found this in astroid 2.12.5 and pylint 2.15.2 with Python 3.7 and 3.11, but I also tried the latest and greatest (astroid 2.15.5 & pylint 2.17.4), and the same issue is still reproducible.

I guess I should report this bug in astroid itself? Sadly, I have no idea how astroid works, so I am afraid I won't be able to fix this, but it's a start.