🐛 Operation folds are not modifier-aware, leaving modifier body in unexpected state after fold to identity

[J4MMlE]

System and Environment Information

• clang/mlir/llvm version 22.1.0-rc3
• OS: NixOS 26.05 (Yarara), kernel 6.19.10-cachyos, x86_64
• MQT Core: main (54c3016)

Bug Description

Folds on QCO operations (e.g. POp::fold) are not aware that the op may live inside a modifier body (e.g. qco.ctrl). When a fold reduces the op to identity, the modifier body is left containing only qco.yield - an unexpected state, causing a crash.

Steps to Reproduce

Try to compile this with mqt-cc (will segfault):

module {
    func.func @controlled-zero-phase() {
      %ctrl = qco.alloc : !qco.qubit
      %tgt  = qco.alloc : !qco.qubit
      %cst  = arith.constant 0.0 : f64
      %ctrl_1, %tgt_1 = qco.ctrl(%ctrl) targets(%t = %tgt) {
        %t_1 = qco.p(%cst) %t : !qco.qubit -> !qco.qubit
        qco.yield %t_1
      } : ({!qco.qubit}, {!qco.qubit}) -> ({!qco.qubit}, {!qco.qubit})
      qco.sink %ctrl_1 : !qco.qubit
      qco.sink %tgt_1  : !qco.qubit
      return
    }
  }

[burgholzer]

Thanks for reporting this 👍🏼
This is clearly an oversight. I would think that this was wrong even before the respective simplifications were changed to folds from being canonicalization patterns.

I would guess that the easiest solution is to add another fold for modifier ops that removes such modifier operations if they do not contain any unitaries.
Could you test that?
We definitely want these folds/patterns to also apply within modifiers; so simply excluding these cases does not really work.

[denialhaag]

This is loosely related to #1727.

[denialhaag]

I'm taking care of this together with #1727. 👍