[smaudet]

> The stack traces when something goes wrong are inscrutable under recursion.

Hmm. This is a real issue, for the simple case. If tail recursion is not optimized correctly then you end up with a bunch of stack frames, wasted memory...

I propose partially this is a tooling issue, not a fundamental problem with recursion. For tail recursion the frames could be optimized away and a virtual counter employed.

For more complex cases, I'd argue it matters less. Saving on stack frames is still preferable, however this can be acheived with judicious use of inlining. But the looping construct is worse here, as you cannot see prior invocations of the loop, and so have generally no idea of program execution without resorting to log tracing, while even the inlined recusion will provide some idea of the previous execution path.