Limit for max in nextInt should be 2^31 inclusive, not exclusive

[rtmigo]

Limit for max in nextInt should be 2^31 inclusive, not exclusive

[Quijx]

Also, I would suggest to check if max == 1 << 32 and directly return nextRaw32() in that case, which is more efficient because it saves the modulo and other operations. This might be usefull if a packages uses nextInt on the Random class (so without a dependency on xrandom) to get random bits. Then the user of that package could supply an Xrandom() as a generator and get slightly better performance.

[rtmigo]

That's right, the range limits are fixed in the new release (0.7.1).

[rtmigo]

Also, I would suggest to check if max == 1 << 32 and directly return nextRaw32()

We literally have only one max value, at which such nextInt will work noticeably faster. But there are more than four billion possible max values, at which such a nextInt will become slightly slower.

I have doubts that such an optimization is appropriate. Especially considering that for a particular max == 1 << 32 there is a public nextRaw32 method. And the nextRaw32 is apparently even faster than an optimized nextInt would be.

Now the user has the opportunity to create a class that inherits Xrandom and overload the nextInt method with optimization for nextRaw32. I think it's worth leaving that decision to the user, whether they need it.

[Quijx]

I am currently writing a package that implements an extension on Random which adds a bunch of methods for generating random values of various types, for example BigInt or Uint8List. And since I don't want to add a dependeny to xrandom and the extension should work on any Random, I am using nextInt(1 << 32) a lot to generate random bytes. (And likely others do this too.) But i see that you have added a special case for all powers of two, which is even better. (The two extra & and - operations likely have absolutely negligible impact on performance.)