qml.counts returns all outcomes when requested if MCMs present

doc/releases/changelog-dev.md

@@ -511,6 +511,9 @@ same information.
 
 <h3>Bug fixes 🐛</h3>
 
+*  `qml.counts` returns all outcomes when requested with mid-circuit measurements present.
+  [(#6732)](https://github.com/PennyLaneAI/pennylane/pull/6732)
+
 * `qml.ControlledQubitUnitary` has consistent behaviour with program capture enabled. 
   [(#6719)](https://github.com/PennyLaneAI/pennylane/pull/6719)
 

pennylane/transforms/dynamic_one_shot.py

@@ -385,6 +385,17 @@ def gather_non_mcm(measurement, samples, is_valid, postselect_mode=None):
     """
     if isinstance(measurement, CountsMP):
         tmp = Counter()
+
+        if measurement.all_outcomes:
+            if isinstance(measurement.mv, Sequence):
+                values = [list(m.branches.values()) for m in measurement.mv]
+                values = list(itertools.product(*values))
+                tmp = Counter({"".join(map(str, v)): 0 for v in values})
+            else:
+                values = [list(measurement.mv.branches.values())]
+                values = list(itertools.product(*values))
+                tmp = Counter({float(*v): 0 for v in values})
+
         for i, d in enumerate(samples):
             tmp.update(
                 {k if isinstance(k, str) else float(k): v * is_valid[i] for k, v in d.items()}

tests/measurements/test_counts.py

@@ -396,6 +396,32 @@ def test_counts_binsize(self):
 class TestCountsIntegration:
     # pylint:disable=too-many-public-methods,not-an-iterable
 
+    def test_counts_all_outcomes_with_mcm(self):
+        n_sample = 10
+
+        dev = qml.device("default.qubit", shots=n_sample)
+
+        @qml.qnode(device=dev, mcm_method="one-shot")
+        def single_mcm():
+            m = qml.measure(0)
+            return qml.counts(m, all_outcomes=True)
+
+        res = single_mcm()
+
+        assert list(res.keys()) == [0.0, 1.0]
+        assert sum(res.values()) == n_sample
+
+        @qml.qnode(device=dev, mcm_method="one-shot")
+        def double_mcm():
+            m1 = qml.measure(0)
+            m2 = qml.measure(1)
+            return qml.counts([m1, m2], all_outcomes=True)
+
+        res = double_mcm()
+
+        assert list(res.keys()) == ["00", "01", "10", "11"]
+        assert sum(res.values()) == n_sample
+
     def test_counts_dimension(self):
         """Test that the counts function outputs counts of the right size"""
         n_sample = 10