Collision detection is independent of the view, because the transformations of your objects don't change, you just view them from different perspectives. (So you are still technically colliding in your second example, but you don't see it)
"Technically" is not intuitive.
"View" means "to see".
What I'm seeing is a car stopping without any barrier.
If you look at the code, you will see that only the car is in the view and the wall is not.
Anyone who sees this will find it strange.
"Technically" anything can be done (remedied), but it would be much better if all of this were more intuitive.
But I guess this does not happen with SFML.
What are you trying to achieve anyways? If you just want the wall to be twice as big, why don't you scale it by that factor then?
Indeed, I am already doing this to remedy the situation. It is not just a simple resize, because if the wall has to be relative to the view for the collision to work and I want the wall to be relative to the original window, I have to consider the view offset and redo a series of calculations.
SFML is a great tool, but could be more intuitive.
In any case I'm already reading the book "SFML Essentials" and I think from there I will understand these puzzles.