Bug in action spaces?

[Warry98]

I was checking the code of the environment and noticed that an action wrapper is always used to normalize the actions, the code used for this is:

class NormalizedActions(gym.ActionWrapper):

    def _action(self, action):
        action = (action + 1) / 2
        action *= (self.action_space.high - self.action_space.low)
        action += self.action_space.low
        return action

    def action(self, action_):
        return self._action(action_)

    def _reverse_action(self, action):
        action -= self.action_space.low
        action /= (self.action_space.high - self.action_space.low)
        action = action * 2 - 1
        return action

But even though this wrapper is used, I believe that the action space doesn't get updated to account it, this causes the action limits to end up being the original ones instead of [-1, 1] which could cause compatibility problems with certain implementations in some of the scenarios.