Combine circuits with clbits onto those with none

[nonhermitian]

fixes #6816

Summary

Allows for the common situation where one combines circuits with measurements (clbits) to those without. Currently you can do this by reversing the order of combine but that is not intuitive at all. Given that many (if not the vast majority) workflows involve the situation described in #6816, it makes sense to allow it. It also requires minimal modifications to do so.

Details and comments

[jakelishman]

Sorry this took a while. This generally looks fine to me as a special case utility. Please could you add a "feature" release note as well, and a test of this compose path with registerless clbits?

[nonhermitian]

@jakelishman I think this is done now. Incorporating registerless bits was a bit tricky because they can be arbitrarily mixed and matched with registers. As there is no (not deprecated or raising) way of querying if a bit has a register, I basically had to resort to looping over bits to see if they are in any registers.

[jakelishman]

As there is no (not deprecated or raising) way of querying if a bit has a register

There is now, though it wasn't added when the original registerless bits implementation was added for some reason. You can do qc.find_bit(bit).registers to find all tuples of register, index that the bit is in. In this case, though, the extra logic isn't necessary: add_register already does it, and won't add bits that are already in the circuit. The example suggestion I made inline should have worked verbatim, even with bits that are in registers (if it didn't, it's a bug).

[nonhermitian]

add_register already does it, and won't add bits that are already in the circuit.

add_register will raise if given a register that already exists in the circuit. So I can't just loop over bits and add to clbits or do add_register. This is is basically why there are two loops in my solution; I need to see if a bit is registerless or not, and make sure I didn't already add the register when looking at earlier bits.

[jakelishman]

add_register raises if a register is already in a circuit, but add_bit does not add any containing register, nor should it, because bits can be in multiple registers, and registers don't own bits. The solution I wrote should have worked verbatim. For example:

In [8]: from qiskit.circuit import QuantumCircuit, QuantumRegister, ClassicalRegister
   ...: qr = QuantumRegister(5)
   ...: cr = ClassicalRegister(5)
   ...: qc = QuantumCircuit()
   ...: qc.add_bits(qr[0:3])
   ...: qc.add_bits(cr[2:4])
   ...: qc.qregs, qc.cregs
Out[8]: ([], [])

In [9]: qc.qubits, qc.clbits
Out[9]:
([Qubit(QuantumRegister(5, 'q3'), 0),
  Qubit(QuantumRegister(5, 'q3'), 1),
  Qubit(QuantumRegister(5, 'q3'), 2)],
 [Clbit(ClassicalRegister(5, 'c3'), 2), Clbit(ClassicalRegister(5, 'c3'), 3)])

In [10]: qc.add_register(qr)
    ...: qc.add_register(cr)
    ...: # Some of the bits from these registers are already in the circuit,
    ...: # but neither of these calls raise.
    ...: qc.qregs, qc.cregs
Out[10]: ([QuantumRegister(5, 'q3')], [ClassicalRegister(5, 'c3')])

In [11]: qc.qubits, qc.clbits
Out[11]:
([Qubit(QuantumRegister(5, 'q3'), 0),
  Qubit(QuantumRegister(5, 'q3'), 1),
  Qubit(QuantumRegister(5, 'q3'), 2),
  Qubit(QuantumRegister(5, 'q3'), 3),
  Qubit(QuantumRegister(5, 'q3'), 4)],
 [Clbit(ClassicalRegister(5, 'c3'), 2),
  Clbit(ClassicalRegister(5, 'c3'), 3),
  Clbit(ClassicalRegister(5, 'c3'), 0),
  Clbit(ClassicalRegister(5, 'c3'), 1),
  Clbit(ClassicalRegister(5, 'c3'), 4)])

If we needed the extra checks, the QuantumCircuit.find_bit function I mentioned above can be used to avoid the quadratic complexity of the nested for-loops, but this is effectively already what add_register does.

[nonhermitian]

Ok yeah, I was confused I guess. I updated the code.

[jakelishman]

No worries - thanks for the tests Paul, it looks good, and it's a nice convenience feature.