cc. @eendebakpt, @nineteendo

Could you expand on the motivation for removing this news entry?

We ask that all changes that will impact on users have a news entry attached to them. It seems to me that a 6% speedup for such a widely used function is a significant improvement that will be appreciated by users, who will therefore appreciate seeing a mention of it in the changelog.

I am fine with either removing the news entry (the performance gain is real, but not huge), or making it more precise.

Looking at the PR there is also a performance improvement for the join in posixpath.py, so not only windows. The number 6% is debatable: for many arguments of join the gain is smaller, for few arguments the gain is larger. And the gain depends on the platform (and probably also on the input type str/bytes and whether the paths are absolute or not). So I would suggest something neutral like

Improve performance of :func:`os.path.join`.

Any user who wants to know the details can lookup the PR.

Yeah, that's certainly fair, "6%" is probably a bit misleading! Arguably a "false sense of precision" rather than "imprecision" ;)

Could you expand on the motivation for removing this news entry?

Oops, it's (implicitly given) in the commit message :)

The speedup given in the PR was:

For Windows, between 5% and 14% faster.
For *nix, between 0% and 35% faster.

Improve performance of :func:os.path.join.

Yep, I would be fine with a vaguer formulation.

Arguably a "false sense of precision" rather than "imprecision" ;)

"FAKE precision!" 🧢 😆

Competing PR:

• gh-117648: Amend NEWS entry #117697

Closed in favour of #117697 🔨

I'm incredibly sorry for all the caused trouble. According to my benchmark the performance increase was around 6% on Windows.
And on Unix, we only avoided the overhead of the map() call, which is not noticeable for a large number of components.
Should performance numbers only be mentioned in the news entry when above 100%?

I'm incredibly sorry for all the caused trouble.

No worries; we can amend NEWS entries.

Should performance numbers only be mentioned in the news entry when above 100%?

I think there is no clear answer; it depends on the API, the use-case, and the performance impact. Sometimes a vague wording is fine. Other times, a more precise wording is warranted.