[mypyc] Fixing index variable in for-loop with builtins.enumerate. (#18202)

Fixes [mypyc/mypyc#1046](mypyc/mypyc#1046)

This change fixes two problems:
1. The index variable was getting instantiated even while enumerating an
empty iterable.
2. After exiting the for-loop, the value of the index variable is off by
1 (see issue linked above).

This change fixes both problems by assigning the temporary register to
the index variable at the beginning of the for-loop body. Before this
change, this assignment was happening before the for-loop and at the end
of the for-loop body.

Author: advait-dixit

mypyc/irbuild/for_helpers.py

@@ -985,7 +985,6 @@ def init(self) -> None:
         zero = Integer(0)
         self.index_reg = builder.maybe_spill_assignable(zero)
         self.index_target: Register | AssignmentTarget = builder.get_assignment_target(self.index)
-        builder.assign(self.index_target, zero, self.line)
 
     def gen_step(self) -> None:
         builder = self.builder
@@ -997,7 +996,9 @@ def gen_step(self) -> None:
             short_int_rprimitive, builder.read(self.index_reg, line), Integer(1), IntOp.ADD, line
         )
         builder.assign(self.index_reg, new_val, line)
-        builder.assign(self.index_target, new_val, line)
+
+    def begin_body(self) -> None:
+        self.builder.assign(self.index_target, self.builder.read(self.index_reg), self.line)
 
 
 class ForEnumerate(ForGenerator):

mypyc/test-data/irbuild-statements.test

@@ -864,33 +864,31 @@ def g(x: Iterable[int]) -> None:
 [out]
 def f(a):
     a :: list
-    r0 :: short_int
-    i :: int
-    r1 :: short_int
+    r0, r1 :: short_int
     r2 :: native_int
     r3 :: short_int
     r4 :: bit
+    i :: int
     r5 :: object
     r6, x, r7 :: int
     r8, r9 :: short_int
 L0:
     r0 = 0
-    i = 0
     r1 = 0
 L1:
     r2 = var_object_size a
     r3 = r2 << 1
     r4 = int_lt r1, r3
     if r4 goto L2 else goto L4 :: bool
 L2:
+    i = r0
     r5 = CPyList_GetItemUnsafe(a, r1)
     r6 = unbox(int, r5)
     x = r6
     r7 = CPyTagged_Add(i, x)
 L3:
     r8 = r0 + 2
     r0 = r8
-    i = r8
     r9 = r1 + 2
     r1 = r9
     goto L1
@@ -900,25 +898,23 @@ L5:
 def g(x):
     x :: object
     r0 :: short_int
-    i :: int
     r1, r2 :: object
-    r3, n :: int
+    i, r3, n :: int
     r4 :: short_int
     r5 :: bit
 L0:
     r0 = 0
-    i = 0
     r1 = PyObject_GetIter(x)
 L1:
     r2 = PyIter_Next(r1)
     if is_error(r2) goto L4 else goto L2
 L2:
+    i = r0
     r3 = unbox(int, r2)
     n = r3
 L3:
     r4 = r0 + 2
     r0 = r4
-    i = r4
     goto L1
 L4:
     r5 = CPy_NoErrOccured()

mypyc/test-data/run-loops.test

@@ -228,6 +228,7 @@ def nested_enumerate() -> None:
             assert i == inner
             inner += 1
         outer += 1
+    assert i == 2
     assert outer_seen == l1
 
 def nested_range() -> None:
@@ -465,6 +466,29 @@ assert g([6, 7], ['a', 'b']) == [(0, 6, 'a'), (1, 7, 'b')]
 assert f([6, 7], [8]) == [(0, 6, 8)]
 assert f([6], [8, 9]) == [(0, 6, 8)]
 
+[case testEnumerateEmptyList]
+from typing import List
+
+def get_enumerate_locals(iterable: List[int]) -> int:
+    for i, j in enumerate(iterable):
+        pass
+    try:
+        return i
+    except NameError:
+        return -100
+
+[file driver.py]
+from native import get_enumerate_locals
+
+print(get_enumerate_locals([]))
+print(get_enumerate_locals([55]))
+print(get_enumerate_locals([551, 552]))
+
+[out]
+-100
+0
+1
+
 [case testIterTypeTrickiness]
 # Test inferring the type of a for loop body doesn't cause us grief
 # Extracted from somethings that broke in mypy