Remove `warn__inc_set_subtensor1` flag

aesara/configdefaults.py

@@ -1454,18 +1454,6 @@ def add_deprecated_configvars():
         in_c_key=False,
     )
 
-    config.add(
-        "warn__inc_set_subtensor1",
-        (
-            "Warn if previous versions of Aesara (before 0.7) could have "
-            "given incorrect results for inc_subtensor and set_subtensor "
-            "when using some patterns of advanced indexing (indexing with "
-            "one vector or matrix of ints)."
-        ),
-        BoolParam(_warn_default("0.7")),
-        in_c_key=False,
-    )
-
     config.add(
         "warn__round",
         "Warn when using `tensor.round` with the default mode. "

aesara/tensor/subtensor.py

 import logging
 import sys
-import warnings
 from itertools import chain, groupby
 from textwrap import dedent
 from typing import Iterable, List, Tuple, Union
@@ -1337,22 +1336,6 @@ def inc_subtensor(
             if v != "x" and (v - dim_offset) >= 0:
                 y_order[v - dim_offset] = i
 
-        # Warn if this code path would have produced wrong results in the past
-        if config.warn__inc_set_subtensor1:
-            # Dimshuffle pattern for y that would be equivalent to past code
-            prev_y_order = ["x"] * (dim_offset) + list(range(y.ndim))
-            if y_order != prev_y_order:
-                warnings.warn(
-                    "Although your current code is fine, please note that "
-                    "earlier versions prior to 0.7 (or this development "
-                    "version) may have yielded an incorrect result in "
-                    "this `inc_subtensor` or `set_subtensor` operation. "
-                    "To remove this warning, you can either set the "
-                    "`warn__inc_set_subtensor1` config option to `False`, "
-                    'or `warn__ignore_bug_before` to at least "0.7".',
-                    stacklevel=2,
-                )
-
         inner_incsubtensor = inc_subtensor(
             inner_x,
             y.dimshuffle(y_order),
@@ -1385,20 +1368,6 @@ def inc_subtensor(
         else:
             flattened_y = y
 
-        # Warn if this code path would have produced wrong results in the past
-        if config.warn__inc_set_subtensor1:
-            if inner_x.ndim > 1 and sum(y.broadcastable) > 0:
-                warnings.warn(
-                    "Although your current code is fine, please note that "
-                    "earlier versions prior to 0.7 (or this development "
-                    "version) may have yielded an incorrect result in "
-                    "this `inc_subtensor` or `set_subtensor` operation. "
-                    "To remove this warning, you can either set the "
-                    "`warn__inc_set_subtensor1` config option to `False`, "
-                    'or `warn__ignore_bug_before` to at least "0.7".',
-                    stacklevel=2,
-                )
-
         inner_incsubtensor = inc_subtensor(
             inner_x,
             flattened_y,

tests/tensor/test_subtensor.py

@@ -1362,31 +1362,24 @@ class TestSubtensor(utt.OptimizationTestMixin):
         shape_i = ((4,), (4, 2))
         shape_val = ((3, 1), (3, 1, 1))
 
-        # Disable the warning emitted for that case
-        orig_warn = config.warn__inc_set_subtensor1
-        try:
-            config.warn__inc_set_subtensor1 = False
-
-            for i, shp_i, shp_v in zip(sym_i, shape_i, shape_val):
-                sub_m = m[:, i]
-                m1 = set_subtensor(sub_m, np.zeros(shp_v))
-                m2 = inc_subtensor(sub_m, np.ones(shp_v))
-                f = aesara.function([m, i], [m1, m2], mode=self.mode)
-
-                m_val = random(3, 5)
-                i_val = integers_ranged(min=0, max=4, shape=shp_i)
-                m1_ref = m_val.copy()
-                m2_ref = m_val.copy()
-
-                m1_val, m2_val = f(m_val, i_val)
-                for idx in i_val.ravel():
-                    m1_ref[:, idx] = 0
-                    m2_ref[:, idx] += 1
-
-                assert np.allclose(m1_val, m1_ref), (m1_val, m1_ref)
-                assert np.allclose(m2_val, m2_ref), (m2_val, m2_ref)
-        finally:
-            config.warn__inc_set_subtensor1 = orig_warn
+        for i, shp_i, shp_v in zip(sym_i, shape_i, shape_val):
+            sub_m = m[:, i]
+            m1 = set_subtensor(sub_m, np.zeros(shp_v))
+            m2 = inc_subtensor(sub_m, np.ones(shp_v))
+            f = aesara.function([m, i], [m1, m2], mode=self.mode)
+
+            m_val = random(3, 5)
+            i_val = integers_ranged(min=0, max=4, shape=shp_i)
+            m1_ref = m_val.copy()
+            m2_ref = m_val.copy()
+
+            m1_val, m2_val = f(m_val, i_val)
+            for idx in i_val.ravel():
+                m1_ref[:, idx] = 0
+                m2_ref[:, idx] += 1
+
+            assert np.allclose(m1_val, m1_ref), (m1_val, m1_ref)
+            assert np.allclose(m2_val, m2_ref), (m2_val, m2_ref)
 
     def test_adv1_inc_sub_notlastdim_1_2dval_no_broadcast(self):
         # Test that taking 1-dimensional advanced indexing
@@ -1399,34 +1392,27 @@ class TestSubtensor(utt.OptimizationTestMixin):
         shape_i = ((4,), (4, 2))
         shape_val = ((3, 4), (3, 4, 2))
 
-        # Disable the warning emitted for that case
-        orig_warn = config.warn__inc_set_subtensor1
-
-        try:
-            config.warn__inc_set_subtensor1 = False
-            for i, shp_i, shp_v in zip(sym_i, shape_i, shape_val):
-                sub_m = m[:, i]
-                m1 = set_subtensor(sub_m, np.zeros(shp_v))
-                m2 = inc_subtensor(sub_m, np.ones(shp_v))
-                f = aesara.function([m, i], [m1, m2], mode=self.mode)
-
-                m_val = random(3, 5)
-                i_val = integers_ranged(min=0, max=4, shape=shp_i)
-                m1_ref = m_val.copy()
-                m2_ref = m_val.copy()
-
-                m1_val, m2_val = f(m_val, i_val)
-                # We have to explicitly loop over all individual indices,
-                # not as a list or array, numpy only increments the indexed
-                # elements once even if the indices are repeated.
-                for idx in i_val.ravel():
-                    m1_ref[:, idx] = 0
-                    m2_ref[:, idx] += 1
-
-                assert np.allclose(m1_val, m1_ref), (m1_val, m1_ref)
-                assert np.allclose(m2_val, m2_ref), (m2_val, m2_ref)
-        finally:
-            config.warn__inc_set_subtensor1 = orig_warn
+        for i, shp_i, shp_v in zip(sym_i, shape_i, shape_val):
+            sub_m = m[:, i]
+            m1 = set_subtensor(sub_m, np.zeros(shp_v))
+            m2 = inc_subtensor(sub_m, np.ones(shp_v))
+            f = aesara.function([m, i], [m1, m2], mode=self.mode)
+
+            m_val = random(3, 5)
+            i_val = integers_ranged(min=0, max=4, shape=shp_i)
+            m1_ref = m_val.copy()
+            m2_ref = m_val.copy()
+
+            m1_val, m2_val = f(m_val, i_val)
+            # We have to explicitly loop over all individual indices,
+            # not as a list or array, numpy only increments the indexed
+            # elements once even if the indices are repeated.
+            for idx in i_val.ravel():
+                m1_ref[:, idx] = 0
+                m2_ref[:, idx] += 1
+
+            assert np.allclose(m1_val, m1_ref), (m1_val, m1_ref)
+            assert np.allclose(m2_val, m2_ref), (m2_val, m2_ref)
 
 
 def test_take_basic():