What's the real example where you got this? ISTM that if a value can be Any then mypy should shut up about it, not try to be picky. If we consider a type to be a set of values, then Any is the set of all values, so adding strings to it doesn't do anything.

I think Any is the one type that can't be thought of as a set of values. (For example, if we think of Any as the set of all values, then you should only be able to do things to it that can be done to all values... but that's not the case.)

I wasn't directly bitten by this, but I realized this was a bit fishy while I was reviewing #1639.

Hm... The situation in #1639 is a little different. The expression in the except clause must be either an exception or a tuple of exceptions, but it cannot be a tuple of tuples of exceptions. So it is incorrect to treat it as a union, and the code is right to insist that each individual value in the tuple is compatible with BaseException -- not that the tuple as a whole (or the tuple turned into a union) is compatible with BaseException.

I still don't think that from your original example above it follows that Union[X, Any] should be treated differently than Any.

You can still use isinstance(x, X) if you want uses of x to be checked more precisely as an X.

Preserving Any inside a union would potentially let us catch some more errors, but figuring out all the implications of this could be a little tricky. I'd rather not worry about this right now as the current behavior is quite reasonable and it conforms to the spirit of how Any works in general. If we see unions like these in user code with non-trivial frequency then we might want to reconsider.

Let's close this issue then.

I've since seen this issue in user code, and we're more agreed on this being a good idea. Reopening so we don't forget about this.

Should probably also be brought up in the python/typing tracker.

This was implemented a while ago.