Invalid Gregorian dates can be constructed (historical edge case)

[mrec]

When the Gregorian calendar was first introduced, it advanced 10 days against the Julian one, so Thursday 4 October 1582 was followed by Friday 15 October 1582. Octobers 5 through 14 1582 are therefore not valid Gregorian dates, but the library doesn't take account of this.

What I did:

cargo add jiff i.e. default config

use jiff::civil::Date;

fn main() {
    assert!(Date::new(1582, 10, 4).is_ok());
    assert!(Date::new(1582, 10, 5).is_err());
    assert!(Date::new(1582, 10, 14).is_err());
    assert!(Date::new(1582, 10, 15).is_ok());
}

Expected: all asserts succeed

Actual: the two .is_err() asserts fail

To be clear, I don't have a concrete use case for needing this to be handled correctly, and was just curious after reading the Dates That Don't Exist blog post. But the jiff docs state very clearly that

Every Date value is guaranteed to be a valid Gregorian calendar date.

so I think either that guarantee needs qualification or the code needs some special-casing. I'd be happy to attempt a PR, but don't have any history with the project and assume you'll want to take a view first.

[BurntSushi]

Ah yeah this crate uses the proleptic Gregorian calendar. So it behaves as if the Gregorian calendar has always existed.

In other words, the behavior here is correct but the docs need some tweaking. Good call out.

[mrec]

I wonder if it'd be worth defining a constant called INTRODUCED or something, to make "any use of Gregorian dates before this is highly likely to trip people up" sanity checks more obvious.

[BurntSushi]

I don't think that's wise. The Gregorian calendar was not introduced in every locale immediately. For example, Russia was still using the Julian calendar as late as 1917. So there really isn't any one date that one can pinpoint.