Raise a warning/error when a procedure that is marked for inlining in the source code is (mutually-)recursive

[jimbxb]

As the title says.

This should be straightforward to detect--if a proc is in a cyclic SCC and is marked for inlining raise the appropriate error/warning.

Without this in place, the inlining pass, if we ever choose to reach a fixed point (can we ever with recursion?) will loop forever

[pschachte]

Actually, it is possible to inline a mutually recursive proc. Suppose p calls q an q calls p. If you inline the call from p to q, you have a proc p that calls itself. There are cases like this in the test suite (where it infers that it should inline), and the optimiser handles it. Even if p also calls itself this is OK. It's only a problem if q also calls itself.

So there is still a need for a warning, but it's not as simple as just warning about marking a proc in a cyclic SCC for inlining, as I thought at first.

[pschachte]

Inlining of tail recursive procs could be made to work if the inlining is done at the LLVM compilation level, rather than when optimising the LPVM. This would essentially turn a call to a recursive proc into an in-place loop in the caller. The LLVM compiler might just handle this for us if we mark the recursive calls as a tail call and mark the LLVM function for inlining. This could even work for mutual tail recursion, but it would not work for recursive but not tail recursive procs.