Single classical bit conditioning breaks various functions

[TharrmashasthaPV]

Information

• Qiskit Terra version: 0.18.0.dev0+f6b2c43
• Python version: 3.7

What is the current behavior?

The PR #6018 enables classical conditioning of gates on individual bits. However, as mentioned in that PR, this new feature breaks the following functions:

• Condition value validation required. Currently only the non-negativity condition of the condition value is checked. So, passing on any float or Boolean (for creg) also get accepted as a valid condition value which is not quite right. Also, bit length of the condition value is also not checked i.e., passing a m bit integer as the condition value for condition on a creg of n classical bits (where m>n) does not raise error .
• All the drawers break when trying to draw a circuit that contains gates with classical conditioning on a single cbit. ( PR in review. Check Text drawer extended for drawing circuits with single classical bit conditioning #6261 , Latex drawer extended for drawing circuits with single classical bit conditioning #6248 and MPL drawer extended for drawing circuits with single classical bit conditioning #6259 )
• qc.qasm() breaks when circuit qc contains gates with classical conditioning on a single cbit.
• qc.depth() also breaks. Refer to Classical conditioning on individual classical bits now supported #6018 (comment) ( Fixed in Fixed error in qc.depth() when using on circuits with single bit classical conditions #6476 )
• disassemble(qc_assembled) breaks where qc_assembled is an assembled Qobj. (PR in review. Check Fixed diassembler for circuits containing single bit conditioned gates #6989 .)
• qc.num_connected_components breaks. ( Fixed in Fixed qc.num_connected_components() for circuits with bit conditioned gates #6564.)
• circuit_to_instruction(qc) breaks. (PR in review Fixes circuit_to_instruction() method for circuits containing single bit conditions #6537.)
• _check_wires_list() and substitute_node_with_dag() methods in qiskit/dagcircuit/dagcircuit.py break.
• _is_same_c_conf() method in ForwardMatch and BackwardMatch classes in qiskit/transpiler/passes/optimization/template_matching breaks.
• run() method in ConsolidateBlocks class in qiskit/transpiler/passes/optimization breaks.
• qpy serialization support (PR in review Handle single bit conditions in QPY #6775)

Suggested solutions

The fixes should be simple. Need to look for parts that have ClassicalRegister in condition in these functions and extend them also to cases when Clbit is in condition.

[mtreinish]

I just added qpy serialization support to the list of things here. It was pointed out to me that trying to dump (and likely load too) a circuit with a single bit condition won't serialize correctly and error. The source is this line:

https://github.com/Qiskit/qiskit-terra/blob/main/qiskit/circuit/qpy_serialization.py#L699

because a Clbit object doesn't have a name. We might need a qpy format v2 to adjust the condition representation in the serialization format (https://github.com/Qiskit/qiskit-terra/blob/main/qiskit/circuit/qpy_serialization.py#L177-L179 ) to handle the lack of a register. Although, thinking out loud we may be able to hack it in by setting the name to be a the str(index) of the clbit in the circuit since that is not a valid register name.

[jakelishman]

As I understand it, QASM 2 doesn't support conditional operations on anything other than a complete register, so the circuit

import qiskit.circuit.qpy_serialization
from qiskit.circuit import QuantumRegister, QuantumCircuit, ClassicalRegister, Clbit
qr = QuantumRegister(2)
cr = ClassicalRegister(8)
qc = QuantumCircuit(qr, cr)
qc.measure(0, cr[0])
qc.measure(1, cr[1])
qc.x(0).c_if(cr[0], 1)
qc.x(1).c_if(cr, 3)

can't be translated into valid QASM 2. I had intended to go around this by emitting additional 1-bit classical registers and copying the data, but there's also no classical assignment operation, so that doesn't work either.

In theory it is possible to emit valid QASM if the only classical bits used aren't attached to registers, though - instead of binding them all into a single creg regless[...]; we separate them. I don't know if we want to put the effort into supporting this if we can't also support subscripts of ClassicalRegister?

I suspect the "best" solution to this within the framework of QASM 2 is just to emit QasmError in this case? It doesn't seem very satisfying, though.

[HuangJunye]

@TharrmashasthaPV we are planning to release 0.20 soon. Are you planning to fix the rest of issues listed here?