The master benchmark:
sys.gettotalrefcount: 14779
sys.gettotalrefcount: 19024
sys.gettotalrefcount: 23269
sys.gettotalrefcount: 27514
sys.gettotalrefcount: 31759
sys.gettotalrefcount: 36004
sys.gettotalrefcount: 40249
sys.gettotalrefcount: 44494
sys.gettotalrefcount: 48739
sys.gettotalrefcount: 52984

After this PR:
sys.gettotalrefcount: 14295
sys.gettotalrefcount: 18056
sys.gettotalrefcount: 21817
sys.gettotalrefcount: 25578
sys.gettotalrefcount: 29339
sys.gettotalrefcount: 33100
sys.gettotalrefcount: 36861
sys.gettotalrefcount: 40622
sys.gettotalrefcount: 44383
sys.gettotalrefcount: 48144

@vstinner Hi, victor, pls take a look after your vacation, thanks.

🤖 New build scheduled with the buildbot fleet by @corona10 for commit 137967a 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

thanks Dong-hee Na for your label ;)

The master benchmark:

What is this benchmark? Which code did you run?

The master benchmark:

What is this benchmark? Which code did you run?

Hm, This should not be called as benchmark. I compare the refleak of this PR with the refleak of master branch. I use debug mode to test it. Because the testcase in https://bugs.python.org/issue1635741#msg355187 use _Py_GetRefTotal()

Since few people care about subinterpreters, I don't think that it's worth it to document this internal change in a NEWS entry: I added "skip news" label.

thanks, victor.