Impossible movements: what's your suggestion?

I think it depends what kind of game you're making. If the position of the player and the things the player can see are important enough that they need to be preserved to prevent cheating or breaking the game, I would consider fading out the view if any part of the player's avatar is intersecting with geometry. I think the player will learn the limitations pretty quickly. You could probably take it one step further and black out the view if the player moves too far from their starting position (in the real world) even if it doesn't lead to intersecting.

In general, if the player is using a traditional control method to move their avatar around in the world, that should be the only thing that results in a permanent change in position in-game. Everything else should be treated like leaning (or stretching), to the extent that you allow it.

My feeling is that allowing a hybrid of traditional controls and actual body movement to move the avatar (aside from leaning) would only confuse and disorient the player anyway.

If you're using real-world body movement rather than traditional control methods to move the avatar, that changes things. I don't have any answers for that at the moment...

Thanks for the reply. I only just recently started experimenting with real work walking (stepping is better word, because of the wires). The reason for this is that it seems natural extension when having implemented full body movements. Maybe the best solution is just not do real world stepping?

Another option is to have the rule not to include situations like this in the scenes, because there is no way of handling it properly.

Then my ideas when you do want to do things like this. I tend to base my algorithms on basic rules:
A. black out when you eyes enter solid objects, do not black out when they are not
B. do not allow any virtual body movement when the screen is blacked out.
C. Other body parts cannot enter solid objects.
D. Try to follow the real body movements as closely as possible with the highest priority for the head.
E. If the trackers go to a position which is not possible with the virtual body: try to recalibrate.

If I apply these rules to the situations I end up with this:
1 The table stops your legs (rule C), but the head keeps moving forward (rule D). However it is not possible to follow the horizontal forward movement of the head without letting the virtual body do strange movements like lifting the feet from the floor. Therefore the best I can do is have the upper body bend forward in the direction of where the head should be. This will result that the head will move downward and will slowly stop moving forward. This is a movement which does not match the real world movements of the HMD, but we are doing something that the player should not do in the first place, so he or she is allowed to suffer a bit...
I am still wondering what result rule E will have in this case. I guess I have to try to see it.

2. Based on rule A, the screen blacks out every time your eyes enter the beam. Rule B gives us that your body does not move in those situations so you can only exit the beam at the point where it entered. Thus it is not possible to move through the beam in any way.

3. When standing up, the system detects an impossible movement: your head should move higher than it possibly can. Something must be wrong, so it starts a recalibration (rule E). The result is that both you and your avatar are standing up. However if you sit down again, no recalibration is necessary: both you and you avatar will be sitting or crouching. But this is not what the player wanted: he wanted to play on with him sitting and the avatar standing. For this, he/she should start recalibration manually: the avatar will go to the starting position (standing up) and will match the current positions of the trackers with this position.

What are your opinions on the basic rules and how they work out?