Fixed #24: Fixed edge case issue where constraint "> 0" does not hand…

[mattfarina]

…le "0.0.1-alpha" properly.

[sdboyer]

so when i'd toyed with doing this before, i did something similar, except that i had a uint8 field that i used to express a three-way distinction - zero, infinity, or "normal" version. here, you've just got the two-way distinction. this ends up being important for handling ranges correctly.

as long as this general problem is being dealt with here, perhaps it makes sense to handle all three?

[mattfarina]

@sdboyer I kept this entirely internal on purpose. I expected it will need to change. The reason I didn't handle a 3 way state out of the box (since we've talked about this before) is I couldn't come up with a use case for infinity. I expected it was my own lack of imagination at that time but it meant I couldn't write tests for infinity. Can you share a use case?

[sdboyer]

sure - infinity is the "value" that a range has if it's got no upper bound.

[sdboyer]

to be clear, i'm not sure i'm fully happy with having the three-way distinction, either. i'm gonna have a PR soon (today, i think) against 2.x that has an implementation, though, so we can have a look!

[mattfarina]

@sdboyer I moved the checking logic to the constraint checking rather than the version instances. It seemed cleaner. And, there's no need to store this state.