Selecting values for usize

[russel]

I have a test with signature:

fn a_test(n: usize) -> bool

but all the selected values of n are in (0, 100) which seems counter-intuitive. Is this what I should expect?

[BurntSushi]

It is what the current implementation defaults to, yes. It is not ideal. A better generation strategy is welcome.

You can however change the range at which usize numbers are generated by creating a StdGen with a custom size and setting it on QuickCheck.

[DurhamG]

@BurntSushi If we have to create a QuickCheck object and set a new generator on it, does this mean we won't be able to use the quickcheck! macro?

I have to admit, it's pretty absurd that a fuzz testing library only checks numbers <100. We had a bug in some serialization/deserialization code of u64's and it was never invoked because it never received a number that required more than one byte to serialize.

[BurntSushi]

If we have to create a QuickCheck object and set a new generator on it, does this mean we won't be able to use the quickcheck! macro?

Yes.

it's pretty absurd

You might reconsider your approach. This is unconstructive complaining that I absolutely refuse to deal with.

[DurhamG]

You're right, it wasn't helpful. Sorry, I was just frustrated.

I see someone proposed a different distribution here. Any thoughts on that style of solution? I can look into making a pull request if that approach sounds reasonable.

[BurntSushi]

@DurhamG Ah yeah! That actually sounds like a great approach. We do the same thing for strings IIRC, so doing it for numbers makes a lot of sense. That's probably why I picked such a small range initially, to cover special cases that are typically small numbers (e.g., 0 and 1). But moving that to a more sophisticated distribution that includes the full space of integers sounds great to me.

[plul]

Do you have an update on this? #221 was closed without merging, but I cannot see why.

[BurntSushi]

I didn't close it. Someone is welcome to resubmit it.