GeezerVison

🤖 New build scheduled with the buildbot fleet by @iritkatriel for commit 6504e9b 🤖

If you want to schedule another build, you need to add the 🔨 test-with-refleak-buildbots label again.

Looking at the logic and the test changes I wonder if this is a good idea?

The example on the issue looks pretty convincing: #103590 (comment)

The example on the issue looks pretty convincing: #103590 (comment)

I find that comment a bit confusing, but a later comment by Carl is indeed convincing.

LGTM.

Should this be backported to 3.11?

Should this be backported to 3.11?

I thought so, but then I got the impression that you and @ambv thought it shouldn't.

So I'm changing my mind, there can't be too much code out there that would be broken by this fix. What does @ambv think?

This is a bug. I agree there is probably little code out there that will be affected by the difference.

However, it's somewhat of a subtle change so user code is unlikely to be tested against it, and if a difference is experienced, the error displayed might be surprising. I'm thinking: "it works on my local box but on CI there's an exception, both are 3.11". While we continue improving 3.11, some installations will be forever stuck at 3.11.1 or so.

But given 3.11's low adoption so far I tend to agree that backporting is probably harmless and ultimately more helpful.

Let's backport then. But let's also be explicit about the change in the docs then. 3.12 whatsnew suggests this change only made it to 3.12, we need to amend that with a mention like "also changed in 3.11.4", and mention this in the except-star docs for both releases.

Good plan. Thanks!