Bug report

Bug description:

For my Python compiler Nuitka, I use CPython as the oracle of what the correct behaviour is. I am running some tests that I used to clarify the behavior from decades ago, in this case I wanted to know when exactly the iterator creation is used. I striped this test a bunch, so the regression is still visible. I first noticed the issue on GitHub Actions, where 3.13.0 got replaced with 3.13.1 for Windows and Linux, but it applies to all OSes. See below for a diff, that the same iterator is created multiple times.

""" Generator expression tests

"""
from __future__ import print_function
import inspect
print("Generator expression that demonstrates the timing:")
def iteratorCreationTiming():
    def getIterable(x):
        print("Getting iterable", x)
        return Iterable(x)
    class Iterable:
        def __init__(self, x):
            self.x = x  # pylint: disable=invalid-name
            self.values = list(range(x))
            self.count = 0
        def __iter__(self):
            print("Giving iterator now", self.x)
            return self
        def __next__(self):
            print("Next of", self.x, "is", self.count)
            if len(self.values) > self.count:
                self.count += 1
                return self.values[self.count - 1]
            else:
                print("Raising StopIteration for", self.x)
                raise StopIteration
        # Python2/3 compatibility.
        next = __next__
        def __del__(self):
            print("Deleting", self.x)
    gen = ((y, z) for y in getIterable(3) for z in getIterable(2))
    print("next value is", next(gen))
    res = tuple(gen)
    print("remaining generator is", res)
    try:
        next(gen)
    except StopIteration:
        print("Usage past end gave StopIteration exception as expected.")
        try:
            print("Generator state then is", inspect.getgeneratorstate(gen))
        except AttributeError:
            pass
        print("Its frame is now", gen.gi_frame)
    print("Early aborting generator:")
    gen2 = ((y, z) for y in getIterable(3) for z in getIterable(2))
    del gen2
iteratorCreationTiming()

The unified diff between 3.13.0 output (and basically all Python versions before) and 3.13.1 output.


--- out-3.13.0.txt      2024-12-06 12:37:19.447115100 +0100
+++ out-3.13.1.txt      2024-12-06 12:37:23.452239500 +0100
@@ -1,9 +1,11 @@
 Generator expression that demonstrates the timing:
 Getting iterable 3
 Giving iterator now 3
+Giving iterator now 3
 Next of 3 is 0
 Getting iterable 2
 Giving iterator now 2
+Giving iterator now 2
 Next of 2 is 0
 next value is (0, 0)
 Next of 2 is 1
@@ -13,6 +15,7 @@
 Next of 3 is 1
 Getting iterable 2
 Giving iterator now 2
+Giving iterator now 2
 Next of 2 is 0
 Next of 2 is 1
 Next of 2 is 2
@@ -21,6 +24,7 @@
 Next of 3 is 2
 Getting iterable 2
 Giving iterator now 2
+Giving iterator now 2
 Next of 2 is 0

The duplicated prints out the iterator creation are new. This is not optimal and new. I don't know if the iterator being through a slot cause cause this or what it is. I checked if generator.c changed but I think it didn't at all.

My self compiled Python 3.13.1 for Linux and the official Windows download agree in behaviour.

Bug report

Bug description:

CPython versions tested on:

Operating systems tested on:

Linked PRs