Tweak Float and Double Arbitrary Instances.

[phadej]

Tries to address comments in https://github.com/nick8325/quickcheck/issues/295 and https://github.com/nick8325/quickcheck/pull/296

Doesn't change how arbitrarySizedFractional works. Now it is only used for (Ratio a) instance.

Also two properties which should fail everytime: the Double generator has a chance of generating same values, and it also hits values which break addition associativity.

---

My take on #295

[cartazio]

Looks like at least a nice starting point. Cc @Bodigrim

[cartazio]

Thx!

[cartazio]

Does this code generate negative zero?

[phadej]

Does this code generate negative zero?

It doesn't. I'll add it.

[cartazio]

I’d also advocate for plus and minus infinity. Since those sometimes have semantic meaning. Ish. And or the smallest and largest magnitude finite floats.

Nans don’t matter / should be runtime exceptions ;)

[phadej]

With infinities you easily get yourself into NaN-land, at which point almost every property is false. That feels like arbitrary choice, who's life you want to make easier/harder

• People testing numeric code (who probably know what they are doing)
• People with some Doubles in their business data types (who probably know less well, to not use Double there).

(1/0.0 :: Double) - (1/0.0 :: Double)
NaN

[cartazio]

Touche. Either way nontrivial chance of zeros or duplicates will go a long way!

For a lotta applied math codes, random all distinct numbers in floating point guarantees the code will work almost too well. :)

On Sat, May 16, 2020 at 4:11 PM Oleg Grenrus notifications@github.com wrote:

With infinities you easily get yourself into NaN-land, at which point almost every property is false. That feels like arbitrary choice, who's life you want to make easier/harder

• People testing numeric code (who probably know what they are doing)
• People with some Doubles in their business data types (who probably know less well, to not use Double there).

(1/0.0 :: Double) - (1/0.0 :: Double)NaN

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/nick8325/quickcheck/pull/297#issuecomment-629699926, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAABBQVALCS7RF3DTFVSOK3RR3XODANCNFSM4NDA2HAQ .

[cartazio]

i emphatically support this change, it seems like it will help MANY people

[phadej]

Resolved conflict.

[phadej]

@nick8325 ping

[nick8325]

Sorry for my extremely slow response to this patch - it looks really useful! Merged now.