Support <,>,==,!=,<=,>=,int,bool,not operations on Constant

dwave/optimization/symbols.pyi

@@ -12,6 +12,8 @@
 #    See the License for the specific language governing permissions and
 #    limitations under the License.
 
+import typing
+
 import numpy.typing
 
 from dwave.optimization.model import Symbol, ArraySymbol
@@ -46,7 +48,22 @@ class BinaryVariable(ArraySymbol):
 
 
 class Constant(ArraySymbol):
-    ...
+    def __bool__(self) -> bool: ...
+    def __index__(self) -> int: ...
+
+    # these methods don't (yet) have an ArraySymbol overload
+    def __ge__(self, rhs: numpy.typing.ArrayLike) -> numpy.typing.NDArray[numpy.bool]: ...
+    def __gt__(self, rhs: numpy.typing.ArrayLike) -> numpy.typing.NDArray[numpy.bool]: ...
+    def __lt__(self, rhs: numpy.typing.ArrayLike) -> numpy.typing.NDArray[numpy.bool]: ...
+
+    @typing.overload
+    def __eq__(self, rhs: ArraySymbol) -> Equal: ...
+    @typing.overload
+    def __eq__(self, rhs: numpy.typing.ArrayLike) -> numpy.typing.NDArray[numpy.bool]: ...
+    @typing.overload
+    def __le__(self, rhs: ArraySymbol) -> LessEqual: ...
+    @typing.overload
+    def __le__(self, rhs: numpy.typing.ArrayLike) -> numpy.typing.NDArray[numpy.bool]: ...
 
 
 class DisjointBitSets(Symbol):

dwave/optimization/symbols.pyx

@@ -20,11 +20,13 @@ import collections.abc
 import json
 import numbers
 
+cimport cpython.object
 import cython
 import numpy as np
 
 from cpython.ref cimport PyObject
 from cython.operator cimport dereference as deref, typeid
+from libc.math cimport modf
 from libcpp cimport bool
 from libcpp.cast cimport dynamic_cast
 from libcpp.optional cimport nullopt, optional
@@ -781,6 +783,12 @@ cdef class Constant(ArraySymbol):
         # Have the parent model hold a reference to the array, so it's kept alive
         model._data_sources.append(array)
 
+    def __bool__(self):
+        if not self._is_scalar():
+            raise ValueError("the truth value of a constant with more than one element is ambiguous")
+
+        return <bool>deref(self.ptr.buff())
+
     def __getbuffer__(self, Py_buffer *buffer, int flags):
         buffer.buf = <void*>(self.ptr.buff())
         buffer.format = <char*>(self.ptr.format().c_str())
@@ -794,6 +802,54 @@ cdef class Constant(ArraySymbol):
         buffer.strides = <Py_ssize_t*>(self.ptr.strides().data())
         buffer.suboffsets = NULL
 
+    def __index__(self):
+        if not self._is_integer():
+            # Follow NumPy's error message
+            # https://github.com/numpy/numpy/blob/66e1e3/numpy/_core/src/multiarray/number.c#L833
+            raise TypeError("only integer scalar constants can be converted to a scalar index")
+
+        return <Py_ssize_t>deref(self.ptr.buff())
+
+    def __richcmp__(self, rhs, int op):
+        # __richcmp__ is a special Cython method
+
+        # If rhs is another Symbol, defer to ArraySymbol to handle the
+        # operation. Which may or may not actually be implemented.
+        # Otherwise, defer to NumPy.
+        # We could also check if rhs is another Constant and handle that differently,
+        # but that might lead to confusing behavior so we treat other Constants the
+        # same as any other symbol.
+        lhs = super() if isinstance(rhs, ArraySymbol) else np.asarray(self)
+
+        if op == cpython.object.Py_EQ:
+            return lhs.__eq__(rhs)
+        elif op == cpython.object.Py_GE:
+            return lhs.__ge__(rhs)
+        elif op == cpython.object.Py_GT:
+            return lhs.__gt__(rhs)
+        elif op == cpython.object.Py_LE:
+            return lhs.__le__(rhs)
+        elif op == cpython.object.Py_LT:
+            return lhs.__lt__(rhs)
+        elif op == cpython.object.Py_NE:
+            return lhs.__ne__(rhs)
+        else:
+            return NotImplemented  # this should never happen, but just in case
+
+    cdef bool _is_integer(self) noexcept:
+        """Return True if the constant encodes a single integer."""
+        if not self._is_scalar():
+            return False
+
+        # https://stackoverflow.com/q/1521607 for the integer test
+        cdef double dummy
+        return modf(deref(self.ptr.buff()), &dummy) == <double>0.0
+
+    cdef bool _is_scalar(self) noexcept:
+        """Return True if the constant encodes a single value."""
+        # The size check is redundant, but worth checking in order to avoid segfaults
+        return self.ptr.size() == 1 and self.ptr.ndim() == 0
+
     @staticmethod
     def _from_symbol(Symbol symbol):
         cdef cppConstantNode* ptr = dynamic_cast_ptr[cppConstantNode](symbol.node_ptr)

releasenotes/notes/constant-comparisons-5258fa0aeb7f435e.yaml

@@ -0,0 +1,6 @@
+---
+features:
+  - Add support for ``<,>,==,!=,<=,>=`` operators between ``Constant`` and array-like objects.
+  - |
+    Add support for ``bool(constant)``, ``int(constant)``, and ``not constant`` when
+    ``constant`` is an instance of ``Constant`` encoding a single scalar value.

tests/test_symbols.py

@@ -15,6 +15,7 @@
 import abc
 import itertools
 import math
+import operator
 import typing
 import unittest
 
@@ -595,6 +596,60 @@ def generate_symbols(self):
         yield A
         yield B
 
+    def test_truthy(self):
+        model = Model()
+
+        self.assertTrue(model.constant(1))
+        self.assertFalse(model.constant(0))
+        self.assertTrue(model.constant(1.1))
+        self.assertFalse(model.constant(0.0))
+
+        self.assertTrue(not model.constant(0))
+        self.assertFalse(not model.constant(1))
+
+        # these are all ambiguous
+        with self.assertRaises(ValueError):
+            bool(model.constant([]))
+        with self.assertRaises(ValueError):
+            bool(model.constant([0, 1]))
+        with self.assertRaises(ValueError):
+            bool(model.constant([0]))
+
+        # the type is correct
+        self.assertIsInstance(model.constant(123.4).__bool__(), bool)
+
+    def test_comparisons(self):
+        model = Model()
+
+        # zero = model.constant(0)
+        one = model.constant(1)
+        onetwo = model.constant([1, 2])
+
+        operators = [
+            operator.eq,
+            operator.ge,
+            operator.gt,
+            operator.le,
+            operator.lt,
+            operator.ne
+        ]
+
+        for op in operators:
+            with self.subTest(op):
+                self.assertEqual(op(one, 1), op(1, 1))
+                self.assertEqual(op(1, one), op(1, 1))
+
+                self.assertEqual(op(one, 2), op(1, 2))
+                self.assertEqual(op(2, one), op(2, 1))
+
+                self.assertEqual(op(one, 0), op(1, 0))
+                self.assertEqual(op(0, one), op(0, 1))
+
+                np.testing.assert_array_equal(op(onetwo, [1, 2]), op(np.asarray([1, 2]), [1, 2]))
+                np.testing.assert_array_equal(op(onetwo, [1, 0]), op(np.asarray([1, 2]), [1, 0]))
+                np.testing.assert_array_equal(op([1, 2], onetwo), op(np.asarray([1, 2]), [1, 2]))
+                np.testing.assert_array_equal(op([1, 0], onetwo), op(np.asarray([1, 0]), [1, 2]))
+
     def test_copy(self):
         model = Model()
 
@@ -604,6 +659,21 @@ def test_copy(self):
         np.testing.assert_array_equal(A, arr)
         self.assertTrue(np.shares_memory(A, arr))
 
+    def test_index(self):
+        model = Model()
+
+        self.assertEqual(list(range(model.constant(0))), [])
+        self.assertEqual(list(range(model.constant(4))), [0, 1, 2, 3])
+
+        with self.assertRaises(TypeError):
+            range(model.constant([0]))  # not a scalar
+
+        self.assertEqual(int(model.constant(0)), 0)
+        self.assertEqual(int(model.constant(1)), 1)
+
+        with self.assertRaises(TypeError):
+            int(model.constant([0]))  # not a scalar
+
     def test_noncontiguous(self):
         model = Model()
         c = model.constant(np.arange(6)[::2])