I question the validity of 2. A properly implemented Fisher-Yates will not require generating random numbers with min=max. If you need this case, it implies that you're doing an additional superfluous iteration.

A properly implemented Fisher-Yates

Have you seen how most PHP devs write code? Demanding proper
implementations is literally blaming the implementor, which is the same
error in reason that allowed the Sony ECDSA failure to occur.

On Jul 13, 2015 6:00 PM, "Nikita Popov" notifications@github.com wrote:

I question the validity of 2. A properly implemented Fisher-Yates will
not require generating random numbers with min=max. If you need this case,
it implies that you're doing an additional superfluous iteration.

—
Reply to this email directly or view it on GitHub.

I don't understand your analogy. If you get an exception due to an off-by-one error in your shuffling implementation, you will be forced to improve your implementation. Why do you consider that stricter input validation should result in broken crypto-system implementations?

I'm saying random_int() should behave like mt_rand() so we can encourage
ease of adoption.

http://3v4l.org/WsfED
On Jul 13, 2015 6:14 PM, "Nikita Popov" notifications@github.com wrote:

I don't understand your analogy. If you get an exception due to an
off-by-one error in your shuffling implementation, you will be forced to
improve your implementation. Why do you consider that stricter input
validation should result in broken crypto-system implementations?

—
Reply to this email directly or view it on GitHub
#1398 (comment).

That's an argument I can buy ;)

FYI: These changes have also been discussed on the internals mailing list.

Closing this since #1397 now has all the newly-agreed upon logic.