This is less readable than before, but maybe we have to accept these changes to have the benefits of ruff. Renaming the i would help here though. Another option would be to turn of the performance rules for the tests (as test performance is typically not critical)

It's very much out-of-scope here, but generally speaking it's preferable to use parametric tests instead of putting assertions in a for-loop 🤷🏻

But even so, I agree that it wouldn't hurt to rename i to something more descriptive.

To tell the truth, the range(len()) → enumerate() change has not been suggested by ruff. I added it for consistency with the desired.keys() → desired.items() change a few lines before.

I am happy to revert the range(len)) → enumerate() change and disable the desired.keys() → desired.items(), globally or locally.

Note that PERF rules are not necessarily about "performance" as such. In most cases, these micro-optimisations will have only (very) marginal effect on actual global performance. Rather, the idea of all these rules is to use consistent Python style.

This change is not the one suggested by the latest version of ruff (on my system at least). Instead of the change here we could also write this as

for k in desired.keys():
            if k not in actual:
                raise AssertionError(f"{k} not in {actual}")
            assert_equal(actual[k], desired[k], f'key={k!r}\n{err_msg}')

I seem to recall no fixes were available here, or perhaps only "unsafe fixes". But yes, I went for a different fix to address the "performance" issue, perhaps at the expense of readability. I will follow your advice and use the above change instead.