Unpin required cirq version

[mpharrigan]

Ideally, we'd support any Cirq version >=1.4 <2.0 (i.e. ~=1.4). The breaking change in quantumlib/Cirq#6960 complicates this

[mpharrigan]

we could use cirq.OP_TREE which exists in all versions

[mpharrigan]

The cirq interop simulator seems broken with cirq 1.6

e.g. qualtran/bloqs/arithmetic/addition_test.py:69 (test_add_decomposition[0-0-4])

                except TypeError:
>                   raise TypeError(f"{self.__class__.__name__} doesn't support {op!r}")
E                   TypeError: Simulator doesn't support BloqAsCirqGate(Add)(cirq.LineQubit(0), cirq.LineQubit(1), cirq.LineQubit(2), cirq.LineQubit(3), cirq.LineQubit(4), cirq.LineQubit(5), cirq.LineQubit(6), cirq.LineQubit(7))

../../conda-313/lib/python3.13/site-packages/cirq/sim/simulator_base.py:211: TypeError

[mpharrigan]

tests pass with cirq 1.5.0

[mpharrigan]

mypy is confusing me with cirq 1.5.0. An example error:

qualtran/bloqs/swap_network/cswap_approx.py:75: error: Missing return statement  [return]
qualtran/bloqs/swap_network/cswap_approx.py:82: error: Argument 1 to "list" has incompatible type "Operation | Iterable[OP_TREE]"; expected "Iterable[Operation | Iterable[OP_TREE]]"  [arg-type]
qualtran/bloqs/swap_network/cswap_approx.py:83: error: Argument 1 to "list" has incompatible type "Operation | Iterable[OP_TREE]"; expected "Iterable[Operation | Iterable[OP_TREE]]"  [arg-type]

this worked with OpTree but fails with OP_TREE even though OP_TREE is supposed to include Iterable[OP_TREE] as part of the union

[pavoljuhas]

The mypy should be fixed by #1724.
With respect to the failure in test_add_decomposition I see it as well, for example with

pytest --pyargs "qualtran.bloqs.arithmetic.addition_test::test_add_decomposition[0-0-3]"

I have bisected that failure to quantumlib/Cirq#7537, specifically the addition of if-has_unitary-return-unitary block to cirq.kraus. The problem is that the circuit0 created in the test reports True for cirq.has_unitary(circuit0), but throws TypeError for cirq.unitary(circuit0) after line 79 here -

Qualtran/qualtran/bloqs/arithmetic/addition_test.py

Lines 70 to 79 in 44f95f0

	@pytest.mark.parametrize('a,b,num_bits', itertools.product(range(4), range(4), range(3, 5)))
	def test_add_decomposition(a: int, b: int, num_bits: int):
	num_anc = num_bits - 1
	gate = Add(QUInt(num_bits))
	qubits = cirq.LineQubit.range(2 * num_bits)
	op = gate.on_registers(a=qubits[:num_bits], b=qubits[num_bits:])
	greedy_mm = cirq.GreedyQubitManager(prefix="_a", maximize_reuse=True)
	context = cirq.DecompositionContext(greedy_mm)
	circuit = cirq.Circuit(cirq.decompose_once(op, context=context))
	circuit0 = cirq.Circuit(op)

The exception is the same with Cirq-1.4.0. Here is a patch to trigger it before any calls to simulator:

diff --git a/qualtran/bloqs/arithmetic/addition_test.py b/qualtran/bloqs/arithmetic/addition_test.py
index 3856f7db..94ebe2f6 100644
--- a/qualtran/bloqs/arithmetic/addition_test.py
+++ b/qualtran/bloqs/arithmetic/addition_test.py
@@ -79,2 +79,7 @@ def test_add_decomposition(a: int, b: int, num_bits: int):
     circuit0 = cirq.Circuit(op)
+
+    if cirq.has_unitary(circuit0):
+        cirq.unitary(circuit0)
+    return
+
     ancillas = sorted(circuit.all_qubits())[-num_anc:]

The has_unitary inconsistency happens also for op.gate which gives True for cirq_unitary; the op was created from gate on line 75, but that gate gives False for has_unitary(gate).

@mpharrigan - can you check the code for op.gate._has_unitary_?
This test error should resolve if its _has_unitary_ gave False as the originating gate.

[mpharrigan]

Thanks for the investigative work! It's not a surprise that this was a bug on our side :) I'll take it from here. I have to check our issue tracker as well -- I think I reported a related issue in the past.

[pavoljuhas]

It's not a surprise that this was a bug on our side :)

I am surprised, we have enough bugs to share on our own. :)