Thank you. Now i understand the 2 sources of circularity in the original solution. With my solution they would be gone, as we would be using named subroutines instead of references to a subroutine. Thus, using get_all_text_in() in the hash won't create any leak.

I can now add in the list of pros and cons above:

• + no risk of inducing a memory leak

I guess here that personal preferences will differ.Personally i would go for the "named subroutines" solution, because i can spot recursion errors more easily than memory leaks. And because i like simple logic.

I am interested to hear people's reasons for preferring one or the other.

In fact, the solution with __SUB__ could use a named subroutine, in which case there's no more memory leak risk. Is that right?

We are then left with the pros and cons that i outlined in my first message. And i don't know which i would prefer.