gh-93143: Avoid NULL check in LOAD_FAST based on analysis in the compiler (GH-93144)

Author: sweeneyde

Include/internal/pycore_opcode.h

@@ -406,6 +406,7 @@ def _write_atomic(path, data, mode=0o666):
 
 #     Python 3.12a1 3500 (Remove PRECALL opcode)
 #     Python 3.12a1 3501 (YIELD_VALUE oparg == stack_depth)
+#     Python 3.12a1 3502 (LOAD_FAST_CHECK, no NULL-check in LOAD_FAST)
 
 #     Python 3.13 will start with 3550
 
@@ -419,7 +420,7 @@ def _write_atomic(path, data, mode=0o666):
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3501).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3502).to_bytes(2, 'little') + b'\r\n'
 
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 

Include/opcode.h

@@ -139,12 +139,14 @@ def jabs_op(name, op):
 def_op('COPY', 120)
 def_op('BINARY_OP', 122)
 jrel_op('SEND', 123) # Number of bytes to skip
-def_op('LOAD_FAST', 124)        # Local variable number
+def_op('LOAD_FAST', 124)        # Local variable number, no null check
 haslocal.append(124)
 def_op('STORE_FAST', 125)       # Local variable number
 haslocal.append(125)
 def_op('DELETE_FAST', 126)      # Local variable number
 haslocal.append(126)
+def_op('LOAD_FAST_CHECK', 127)  # Local variable number
+haslocal.append(127)
 jrel_op('POP_JUMP_FORWARD_IF_NOT_NONE', 128)
 jrel_op('POP_JUMP_FORWARD_IF_NONE', 129)
 def_op('RAISE_VARARGS', 130)    # Number of raise arguments (1, 2, or 3)

Lib/importlib/_bootstrap_external.py

@@ -360,7 +360,7 @@ def bug42562():
     -->    BINARY_OP               11 (/)
            POP_TOP
 
-%3d        LOAD_FAST                1 (tb)
+%3d        LOAD_FAST_CHECK          1 (tb)
            RETURN_VALUE
         >> PUSH_EXC_INFO
 
@@ -1399,7 +1399,7 @@ def _prepare_test_cases():
   Instruction(opname='LOAD_CONST', opcode=100, arg=4, argval='I can haz else clause?', argrepr="'I can haz else clause?'", offset=100, starts_line=None, is_jump_target=False, positions=None),
   Instruction(opname='CALL', opcode=171, arg=1, argval=1, argrepr='', offset=102, starts_line=None, is_jump_target=False, positions=None),
   Instruction(opname='POP_TOP', opcode=1, arg=None, argval=None, argrepr='', offset=112, starts_line=None, is_jump_target=False, positions=None),
-  Instruction(opname='LOAD_FAST', opcode=124, arg=0, argval='i', argrepr='i', offset=114, starts_line=11, is_jump_target=True, positions=None),
+  Instruction(opname='LOAD_FAST_CHECK', opcode=127, arg=0, argval='i', argrepr='i', offset=114, starts_line=11, is_jump_target=True, positions=None),
   Instruction(opname='POP_JUMP_FORWARD_IF_FALSE', opcode=114, arg=34, argval=186, argrepr='to 186', offset=116, starts_line=None, is_jump_target=False, positions=None),
   Instruction(opname='LOAD_GLOBAL', opcode=116, arg=3, argval='print', argrepr='NULL + print', offset=118, starts_line=12, is_jump_target=True, positions=None),
   Instruction(opname='LOAD_FAST', opcode=124, arg=0, argval='i', argrepr='i', offset=130, starts_line=None, is_jump_target=False, positions=None),

Lib/opcode.py

@@ -1,5 +1,6 @@
 import dis
 from itertools import combinations, product
+import sys
 import textwrap
 import unittest
 
@@ -682,5 +683,184 @@ def test_bpo_45773_pop_jump_if_false(self):
         compile("while True or not spam: pass", "<test>", "exec")
 
 
+class TestMarkingVariablesAsUnKnown(BytecodeTestCase):
+
+    def setUp(self):
+        self.addCleanup(sys.settrace, sys.gettrace())
+        sys.settrace(None)
+
+    def test_load_fast_known_simple(self):
+        def f():
+            x = 1
+            y = x + x
+        self.assertInBytecode(f, 'LOAD_FAST')
+
+    def test_load_fast_unknown_simple(self):
+        def f():
+            if condition():
+                x = 1
+            print(x)
+        self.assertInBytecode(f, 'LOAD_FAST_CHECK')
+        self.assertNotInBytecode(f, 'LOAD_FAST')
+
+    def test_load_fast_unknown_because_del(self):
+        def f():
+            x = 1
+            del x
+            print(x)
+        self.assertInBytecode(f, 'LOAD_FAST_CHECK')
+        self.assertNotInBytecode(f, 'LOAD_FAST')
+
+    def test_load_fast_known_because_parameter(self):
+        def f1(x):
+            print(x)
+        self.assertInBytecode(f1, 'LOAD_FAST')
+        self.assertNotInBytecode(f1, 'LOAD_FAST_CHECK')
+
+        def f2(*, x):
+            print(x)
+        self.assertInBytecode(f2, 'LOAD_FAST')
+        self.assertNotInBytecode(f2, 'LOAD_FAST_CHECK')
+
+        def f3(*args):
+            print(args)
+        self.assertInBytecode(f3, 'LOAD_FAST')
+        self.assertNotInBytecode(f3, 'LOAD_FAST_CHECK')
+
+        def f4(**kwargs):
+            print(kwargs)
+        self.assertInBytecode(f4, 'LOAD_FAST')
+        self.assertNotInBytecode(f4, 'LOAD_FAST_CHECK')
+
+        def f5(x=0):
+            print(x)
+        self.assertInBytecode(f5, 'LOAD_FAST')
+        self.assertNotInBytecode(f5, 'LOAD_FAST_CHECK')
+
+    def test_load_fast_known_because_already_loaded(self):
+        def f():
+            if condition():
+                x = 1
+            print(x)
+            print(x)
+        self.assertInBytecode(f, 'LOAD_FAST_CHECK')
+        self.assertInBytecode(f, 'LOAD_FAST')
+
+    def test_load_fast_known_multiple_branches(self):
+        def f():
+            if condition():
+                x = 1
+            else:
+                x = 2
+            print(x)
+        self.assertInBytecode(f, 'LOAD_FAST')
+        self.assertNotInBytecode(f, 'LOAD_FAST_CHECK')
+
+    def test_load_fast_unknown_after_error(self):
+        def f():
+            try:
+                res = 1 / 0
+            except ZeroDivisionError:
+                pass
+            return res
+        # LOAD_FAST (known) still occurs in the no-exception branch.
+        # Assert that it doesn't occur in the LOAD_FAST_CHECK branch.
+        self.assertInBytecode(f, 'LOAD_FAST_CHECK')
+
+    def test_load_fast_unknown_after_error_2(self):
+        def f():
+            try:
+                1 / 0
+            except:
+                print(a, b, c, d, e, f, g)
+            a = b = c = d = e = f = g = 1
+        self.assertInBytecode(f, 'LOAD_FAST_CHECK')
+        self.assertNotInBytecode(f, 'LOAD_FAST')
+
+    def test_setting_lineno_adds_check(self):
+        code = textwrap.dedent("""\
+            def f():
+                x = 2
+                L = 3
+                L = 4
+                for i in range(55):
+                    x + 6
+                del x
+                L = 8
+                L = 9
+                L = 10
+        """)
+        ns = {}
+        exec(code, ns)
+        f = ns['f']
+        self.assertInBytecode(f, "LOAD_FAST")
+        def trace(frame, event, arg):
+            if event == 'line' and frame.f_lineno == 9:
+                frame.f_lineno = 2
+                sys.settrace(None)
+                return None
+            return trace
+        sys.settrace(trace)
+        f()
+        self.assertNotInBytecode(f, "LOAD_FAST")
+
+    def make_function_with_no_checks(self):
+        code = textwrap.dedent("""\
+            def f():
+                x = 2
+                L = 3
+                L = 4
+                L = 5
+                if not L:
+                    x + 7
+                    y = 2
+        """)
+        ns = {}
+        exec(code, ns)
+        f = ns['f']
+        self.assertInBytecode(f, "LOAD_FAST")
+        self.assertNotInBytecode(f, "LOAD_FAST_CHECK")
+        return f
+
+    def test_deleting_local_adds_check(self):
+        f = self.make_function_with_no_checks()
+        def trace(frame, event, arg):
+            if event == 'line' and frame.f_lineno == 4:
+                del frame.f_locals["x"]
+                sys.settrace(None)
+                return None
+            return trace
+        sys.settrace(trace)
+        f()
+        self.assertNotInBytecode(f, "LOAD_FAST")
+        self.assertInBytecode(f, "LOAD_FAST_CHECK")
+
+    def test_modifying_local_does_not_add_check(self):
+        f = self.make_function_with_no_checks()
+        def trace(frame, event, arg):
+            if event == 'line' and frame.f_lineno == 4:
+                frame.f_locals["x"] = 42
+                sys.settrace(None)
+                return None
+            return trace
+        sys.settrace(trace)
+        f()
+        self.assertInBytecode(f, "LOAD_FAST")
+        self.assertNotInBytecode(f, "LOAD_FAST_CHECK")
+
+    def test_initializing_local_does_not_add_check(self):
+        f = self.make_function_with_no_checks()
+        def trace(frame, event, arg):
+            if event == 'line' and frame.f_lineno == 4:
+                frame.f_locals["y"] = 42
+                sys.settrace(None)
+                return None
+            return trace
+        sys.settrace(trace)
+        f()
+        self.assertInBytecode(f, "LOAD_FAST")
+        self.assertNotInBytecode(f, "LOAD_FAST_CHECK")
+
+
 if __name__ == "__main__":
     unittest.main()

Lib/test/test_dis.py

@@ -0,0 +1 @@
+Avoid ``NULL`` checks for uninitialized local variables by determining at compile time which variables must be initialized.