My preference is not widely shared. I take a page from Linus Torvalds, define the API to program to, and write a simple wrapper for that API. Then all I ever need to do is write to that wrapper.

I might implement in this case by checking for the existence of the module at run-time, and initialize a bunch of anon subs accordingly. My API wind up behind the scene calling them, and they will automagically do the right thing (even if that thing is to spit out debugging info or else just do nothing).

For the full philosophy, here is a good rant from Linus explaining the why's and wherefore's. Even if you disagree it is a worthwhile read.