Context:

Previously, despite using all_outcomes , qml.counts incorrectly returned only the states that have counts present,

import pennylane as qml
dev = qml.device("default.qubit", shots=10)

@qml.qnode(device=dev, mcm_method="one-shot")
def circuit():
    m1 = qml.measure(0)
    return qml.counts(m1, all_outcomes=True)

>>> circuit()
{0.0: 10}

Description of the Change:

If all_outcomes is specified, the intermediate tmp variable (used to keep track of measurement values from MCMs) defaults to a Counter object that initializes all possible basis states as keys with their values set to zero. This counter is then updated based on the measurement results, where each mcm sample contributes to updating the counts, weighted by a validity factor.

This results in the correct output,

dev = qml.device("default.qubit", shots=10)

@qml.qnode(device=dev, mcm_method="one-shot")
def circuit():
    m1 = qml.measure(0)
    return qml.counts(m1, all_outcomes=True)

>>> circuit()
{0.0: 10, 1.0: 0}
# {0.0: 10} if all_outcomes=False

or,

dev = qml.device("default.qubit", shots=10)

@qml.qnode(device=dev, mcm_method="one-shot")
def circuit():
    m1 = qml.measure(0)
    m2 = qml.measure(1)
    return qml.counts([m1,m2], all_outcomes=True)

>>> circuit()
{'00': 10, '01': 0, '10': 0, '11': 0}
# {'00': 10} if all_outcomes=False

Benefits: all_outcomes kwarg in qml.counts now works as intended when mcm_method="one-shot" and MCM operations are present.

Possible Drawbacks: tmp object creation doesn't scale well with higher qubit counts.

Related GitHub Issues: Fixes #6700

[sc-80206]