This whole bit can be sped up by just replacing it with:

return tuple(dict.fromkeys(params))

Thanks, I will add some tests. By the way, params is a list, tuple(dict.fromkeys(params)) seems not faster than tuple(params), I think it would be better to keep tuple(params):

Sorry, I should have been clearer! I meant replacing this whole list + loop bit with that one line. dict.fromkeys is a nice shortcut to remove duplicates while preserving order.

Ah... I see now. You're trying to avoid the (broken) hashing.

I don't know how I feel about that. I think hashing should be fixed rather than just sidestepped here, since it's so tied to equality. At this point I'm actually in favor of a hybrid approach: combining your fix for __eq__ and the other PR #15400’s fix for __hash__.

@gvanrossum and @ilevkivskyi, thoughts?

@hongweipeng you can revert your most recent commit (3676a50, “code review”). My comment was misguided!

TBH this looks like a hack. If at some point we would make sets ordered the same way as dicts are, it would be good to update the code, but currently I think I like the old way more.