[Question] 'NoneType' object has no attribute 'glfwGetCurrentContext'

[qffof]

Question

Hi！I have some questions for you
background：

gymnasium: 1.0.0
glfw: 2.7.0

I run the code below：

import gymnasium as gym
env=gym.make('Humanoid-v5', render_mode='human')
obs=env.reset()
env.render()

An error was reported ：

  logger.deprecation(
  File "D:\anaconda\envs\Gemma2b\lib\site-packages\gymnasium\envs\mujoco\mujoco_rendering.py", line 359, in __del__
  File "D:\anaconda\envs\Gemma2b\lib\site-packages\gymnasium\envs\mujoco\mujoco_rendering.py", line 352, in free
  File "D:\anaconda\envs\Gemma2b\lib\site-packages\glfw\__init__.py", line 2366, in get_current_context
AttributeError: 'NoneType' object has no attribute 'glfwGetCurrentContext'

When I take the solution given earlier：

import gymnasium as gym
env=gym.make('Humanoid-v5', render_mode='human')
obs=env.reset()
env.render()
breakpoint()

The result of the code run is: The mojoco simulation interface is not responding

[Beanpow]

You may have to close the env before the program ends.

import gymnasium as gym
env=gym.make('Humanoid-v5', render_mode='human')
obs=env.reset()
env.render()
env.close()

[pseudo-rnd-thoughts]

This might be a windows only issue

We don't officially support windows but will try to fix bugs

[Kallinteris-Andreas]

If you're using Windows, try the Linux subsystem.

[a-ayesh]

I ran into this issue as well while using gymnasium to render my MuJoCo environment in Stable-Baselines3. Details and how to replicate are as follows:

Details

OS: Ubuntu 22.04.5 LTS
Python Venv: Anaconda
Python Version: 3.10

How to replicate

from stable_baselines3.common.env_util import make_vec_env
from stable_baselines3.common.evaluation import evaluate_policy
from stable_baselines3 import TD3

# Number of evaluation episodes
N_EVAL_EPISODES = 10

# Environment ID
ENV_ID = 'InvertedPendulum-v5'

# Optimized hyperparameters for the TD3 agent
TD3_OPTIMIZED_HYPERPARAMS = {
    'learning_rate': 0.001,
    'buffer_size': 100000,
    'tau': 0.01,
    'gamma': 0.999,
    'train_freq': 100,
    'gradient_steps': 200,
    'batch_size': 64,
    'policy_delay': 1
}

from lib.utils import *


class TD3Agent:
    """
    TD3 Agent implementation for training and evaluation.
    """

    def __init__(self, env_id: str, hyperparams: dict, verbose: int = 1):
        """
        Initializes the TD3Agent.

        Args:
            env_id (str): The Gym environment ID.
            hyperparams (dict): Dictionary containing hyperparameters for the TD3 agent.
            verbose (int, optional): Verbosity level (default: 1).
        """
        self.env_id = env_id
        self.env = make_vec_env(env_id, n_envs=1, env_kwargs={'render_mode': 'human'})
        self.hyperparams = hyperparams
        self.verbose = verbose
        self.model = None

    def evaluate(self, n_eval_episodes: int = 10) -> tuple:
        """
        Evaluates the TD3 model on the environment.

        Args:
            n_eval_episodes (int): Number of evaluation episodes.

        Returns:
            tuple: A tuple containing mean reward and standard deviation of rewards.
        """
        if self.model is None:
            raise ValueError("Model has not been created or trained yet.")
        mean_reward, std_reward = evaluate_policy(
            self.model,
            self.env,
            n_eval_episodes=n_eval_episodes
        )
        return mean_reward, std_reward

    def load(self, path: str):
        """
        Loads a pre-trained TD3 model from the specified path.

        Args:
            path (str): Path to the saved model file.
        """
        self.model = TD3.load(path, env=self.env)


def run_algorithm():
    """
    Creates and evaluates a TD3 agent using the specified environment and hyperparameters.
    """
    agent = TD3Agent(env_id=ENV_ID, hyperparams=TD3_OPTIMIZED_HYPERPARAMS, verbose=1)
    agent.load("agent")  # Path to the pre-trained agent

    print("Evaluating...")
    mean_reward, std_reward = agent.evaluate(n_eval_episodes=N_EVAL_EPISODES)
    print(f"Stats: Mean Reward: {mean_reward:.2f} +/- {std_reward}\n")

    # agent.env.close() # Comment this to see the error

if __name__ == "__main__":
    run_algorithm()

The above code evaluates a pre-trained TD3 agent using SB3 on the InvertedPendulum-v5 MuJoCo environment. The env component is handled by gymnasium and I was getting the same error as this issue.

What causes the issue

• User doesn't call env.close() to gracefully shutdown the rendering context
• Consequently, when the python program terminates, the dependencies for glfw are unloaded before glfw itself causing it to throw an AttributeError

How to fix this issue

• Simply call env.close() before your python program terminates

[Kallinteris-Andreas]

1. the bug of unresponsive UI when render_mode="human" and breakpoint() is because the execution of the program has paused

import gymnasium

env = gymnasium.make('Humanoid-v5', render_mode='human')
obs, info = env.reset()
# env.render()  # no need to call it with `render_mode="human"`
env.unwrapped.mujoco_renderer.viewer._paused = True  # this pauses the render window, it can be unpaused by pressing the SPACE key
for _ in range(1000):
    env.step(env.action_space.sample())

2. The error is not descriptive to the user and should be fixed Add error message when terminating the MuJoCo renderer without calling env.close #1283

Exception ignored in: <function WindowViewer.__del__ at 0x7f058805ee80>
Traceback (most recent call last):
  File "/home/master-andreas/Gymnasium/gymnasium/envs/mujoco/mujoco_rendering.py", line 377, in __del__
  File "/home/master-andreas/Gymnasium/gymnasium/envs/mujoco/mujoco_rendering.py", line 365, in free
  File "/home/master-andreas/temp_env/lib/python3.13/site-packages/glfw/__init__.py", line 2369, in get_current_context
AttributeError: 'NoneType' object has no attribute 'glfwGetCurrentContext'