Bring back cpRemoveCollisionHandler about chipmunk2d

[slembcke]

There used to be a lot of (really gross) code to implement removing collision handlers in the past. Even after all the work that went into fixing all the known bugs it caused, there were still a lot of really nasty gotchas. For example, removing a handler was actually deferred and could cause nasty dangling pointer issues if you freed your user data object after removing the handler.

I decided a couple years ago that it was a misfeature, and that there really was no way to make it work correctly, sensibly, or safely. The new cpCollisionHandler structs were supposed to fix a lot of that. Since collision pairs reference the handler instead of copying the callback pointers, changes to the handler are immediate with no surprises or dangling pointer issues.

You are unfortunately right that the new struct way has a couple new issues. Setting a callback back to the default is difficult (not really on purpose). I'm not entirely certain how to improve that since the default changes based on if the handler is a regular or wildcard handler.

the persistence of setting things in a returned struct is not obvious, nor memory management/life-cycle issues that come with it

Since you cannot remove a handler, it exists as long as the space does. This is the same as how it used to work if you weren't explicitly removing handlers. This was meant to simplify the (basically undocumented and complicated) memory lifecycle problems with removing handlers in v6. Since Chipmunk only reads from the handlers, whatever values you write into them can be considered permanent.

[ewmailing]

I made manual fixes to the .def file in my branch. It also contains other things that may be of interest to you.
https://github.com/ewmailing/Chipmunk2D/commits/Blurrr7