really drop python<=3.7

[kloczek]

Filter all code over pyupgrade --py38-plus.

[bdarnell]

Most of these actually date back to Python 3.0 or earlier (e.g. the default (object) base class was in 3.0, set literals in 2.7). I'm wary of change for the sake of change but it's probably time to change these and the tool helps make it safer. I've never liked the .format() method (introduced in 2.6), though - I prefer the % operator or f-strings. Looks like pyupgrade has an option to leave string formatting alone. (I wish it would let you run one rule at a time for ease of reviewing - get all of the (object) changes out of the way in one commit, etc)

I'm not sure about that second ruff commit - I think most if not all of those imports are needed for type checking. We already have flake8 that warns about unused imports so there shouldn't really be any.

[kloczek]

flake8 is more and more behind of what ruff is able to detect.
Nevertheless with both commits test suite is OK (tornado.test.runtests) and pytest is failing the same way as without that those changes (as I've reported in #3390)

[bdarnell]

flake8 is more and more behind of what ruff is able to detect.

I'm not in the market for a new linter (well, I may switch from flake8 to plain pyflakes now that black and pep8 are diverging). I care more about a high signal-to-noise ratio than just detecting more stuff (this is why I use pyflakes and not pylint).

Nevertheless with both commits test suite is OK (tornado.test.runtests)

It also needs to pass the linters (tox -e lint. Also tox -e docs although i don't think that's going to be an issue here). The current failure just needs black to be run after pyupgrade, and then we'll see if mypy is happy with the ruff changes.

[kloczek]

OK I've added commit with all changes generated by black.

[bdarnell]

This useless-looking call to encode() is significant - it ensures that the codec is initialized here (where we are presumably executing on the main thread) instead of later in another thread, which has the potential to deadlock.

[bdarnell]

I've got almost all the pyupgrade changes in a new PR, #3403. I had to hack on pyupgrade a bit to disable one change and then break things out into separate commits for easier review. I also found that running pyupgrade in two passes will convert percent formatting to str.format and then str.format to fstrings, which is nice.

[kloczek]

Thank you 👍