@Kwpolska: the failing tests are the baseline. I don't see what went wrong here, though; maybe some dependency changed which now generates a slightly different output? Also, the same failing baseline differences show up for #2594, which has completely different code changes.

That’s a lxml update, I just rebuilt the baseline.

PS. small and obvious fixes do not need to be done through a pull request, you may commit to master directly.

Nope, that doesn't work anymore. I tried that first, and I got an error message that a review is required when trying to push to master. So I had to convert it to a PR first.

Oh, that’s a side effect of requiring reviews and protecting branches. Which does not affect administrators, so I didn’t notice.