Reorder result bits from real devices

[ajavadia]

Fixes #2 (!!)
Fixes #5
Fixes a bunch of other complaints

A current software bug in the remote devices is preventing the format of obtained classical bits from matching what the user expects. This PR addresses that.

With the implementation of qobj across the full stack, this problem will be resolved in the backends themselves. So this is a temporary solution until qobj is in place in all backends.

Description

The reordering is implemented in the ibmqbackend and is easy to turn on/off.
Includes reordering, truncating to match the circuit clbits, and inserting spaces to signify separate classical registers.

Motivation and Context

Numerous bug reports, so I am submitting this even though qobj is close.

How Has This Been Tested?

Wrote 2 tests to check the reordered result against the simulator results, which are correct.

Screenshots (if appropriate):

Example circuit:

q0 = QuantumRegister(3)
q1 = QuantumRegister(2)
q2 = QuantumRegister(1)
c0 = ClassicalRegister(3)
c1 = ClassicalRegister(2)
c2 = ClassicalRegister(1)
qc = QuantumCircuit(q0, q1, q2, c0, c1, c2)
qc.h(q0[0])
qc.cx(q0[0], q0[2])
qc.x(q1[1])
qc.h(q2[0])
qc.barrier()
qc.measure(q0[0], c2[0])
qc.measure(q0[1], c0[1])
qc.measure(q0[2], c1[1])
qc.measure(q1[0], c0[0])
qc.measure(q1[1], c1[0])
qc.measure(q2[0], c0[0])

Output from ibmqx5, before:

Now:

Simulator:

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist:

• My code follows the code style of this project.
• My change requires a change to the documentation.
• I have updated the documentation accordingly.
• I have read the CONTRIBUTING document.
• I have added tests to cover my changes.
• All new and existing tests passed.

[diego-plan9]

Nit: can you use the fully qualified name (ie. import qiskit), for consistency?

[1ucian0]

fixed in b658084

[diego-plan9]

Can you revise these tests? They seem to be raising an Exception intermitently:

qiskit/wrapper/_wrapper.py:150: in execute
    wait, timeout, skip_translation)
qiskit/_compiler.py:77: in execute
    result = backend.run(q_job)
qiskit/backends/ibmq/ibmqbackend.py:125: in run
    _reorder_bits(this_result)  # TODO: remove this after Qobj

    def _reorder_bits(result):
        """temporary fix for ibmq backends.
        for every ran circuit, get reordering information from qobj
        and apply reordering on result"""
        for idx, circ in enumerate(result._qobj['circuits']):
    
            # device_qubit -> device_clbit (how it should have been)
            measure_dict = {op['qubits'][0]: op['clbits'][0]
                            for op in circ['compiled_circuit']['operations']
                            if op['name'] == 'measure'}
    
            res = result._result['result'][idx]
            counts_dict_new = {}
>           for item in res['data']['counts'].items():
E           TypeError: string indices must be integers

when running against ibmqx4 at least. It seems it might be related to a timeout which is not handled properly during run() - the contents of result._result on one of those failures where:

{'backend': 'ibmqx4',
 'job_id': '(redacted)',
 'name': '3ZwDQis6FwX0j5gRpYvmcAEZ273upm',
 'result': 'QISkit Time Out',
 'status': 'ERROR'}

From a practical point of view, they are also a bit problematic: since they depend on a real device, the running time of the tests can be rather large (in my tests, they ranged from 150 to 170 seconds, which is rather large considering that the current running time for all the test suite is ~120 seconds. I'd really like to avoid that one way or the other, by making them opt-in or similar - any ideas?

[1ucian0]

fixed in 5e99aed

[ajavadia]

This should be ready to merge now. Thanks @diego-plan9 for the catch and @1ucian0 for the help.

[diego-plan9]

Thanks @ajavadia and @1ucian0 for fixing the long-standing bugs! 🎉

[msegado]

Thanks for adding this! I'm wondering - will this be released as a bugfix in 0.4.x or as a breaking change in 0.5.x? (Some of my code also does reordering to work around #2, so when this fix lands I'll need to stop doing that to avoid reordering twice!)

[ajavadia]

@msegado we will add this to the 0.5 release, which is close (about a week away).

[msegado]

@ajavadia Got it, thanks for the info!