Hi Paul,

> It seems that your patch is incomplete, as there's a memory leak in the 
> unlikely case that the malloc call fails.

Oops, indeed. Thanks for the review!

> Something like the attached patch instead, perhaps?

Hmm, this patch is freeing the same object through the variable 'dirname'
on one branch and through the variable 'previous_dirname' on the other
branch. A bit hard to understand, and thus likely to introduce more bugs
in the future, I would say. I prefer the attached one. Pushed.

Bruno