Jukka Lehtosalo

Yeah, this is something I left out originally in the interest of simplicity. I don't expect this to have a significant impact on mypy performance, but more generally this can...

A simple implementation would cache the last N created objects and try to reuse them, where N could be, say, 4 or 8. A linear search would be sufficient. A...

This would need to happen during IR building. I don't expect it to be very difficult to implement but not quite trivial either.

### Implementation The main challenge of implementing fixed-width integers seems to be dealing with what we call "error values". Each type currently has a dedicated error value, but this can't...

Initially it would be fine to support only these types: * `i64` * `i32` * `i16` * `u8` Unsigned integers are best avoided when possible, since they wrap around if...

@pranavrajpal A good point about the two use cases. I'm mostly interested in use case 1, since the second one seems to come up much less frequently. Because of the...

I don't like the term "fixed-width integers", since it's long and unwieldy. I'm planning to change the name of this feature to "native integers". I think that this would be...

I'm not very happy with the approach I proposed above for function default values. It can add significant overhead, especially if there are many arguments. Since the vast majority of...

Let's keep the issue also open here for more visibility, since not many users know about https://github.com/mypyc/mypy_mypyc-wheels.

This also affects `len(d)`, if `d` is an instance of a dict subclass. This is not very high priority, since overriding `__len__` is probably very rare, but I'm sure it...