Add seed(InputRange) overload to MersenneTwisterEngine

[jpf91]

Adds a new seed overload as discussed here: http://forum.dlang.org/thread/jr0luj$1ctj$1@digitalmars.com

[dnadlinger]

Maybe add the number of elements required to the error message and documentation?

[dnadlinger]

Also, you might want to add a note to the single integer overload, describing its limitations.

[jpf91]

The the number of elements required is not fixed. It's 624 for Mt19937 but it seems it can be any value for MersenneTwisterEngine. But I added a basic description to the docs.

[dnadlinger]

2^19937 (-1) is only the period for Mt19937, not for MersenneTwisterEngine in general, right?

[jpf91]

Yes, you're right. Removed the 2^19937 note.

[jmdavis]

j is used for indexing mt and comparing against n (which is a size_t), so it should be size_t, not int.

[jpf91]

@jmdavis OK, implemented your suggested changes. rndGen is also using the new seed overload now. However, I had to slightly change isSeedable as well. Without that change, it's not possible to use isSeedable for the new seed overload.

[jpf91]

Yes I should probably do that. Although adding the guarantee about Rng.front was wrong in the first place, but I didn't know that RNGs could be seeded with other types (I implemented isSeedable).

[jpf91]

map doesn't work in CTFE yet. I think we can just leave that as is, as long as rndGen uses the new seed method and we document the problems with the old seed overload.

[jmdavis]

If there's really no relation between the type of front and the type that's used to seed the range, then there should be no check for whether the type of front and the seed type are the same. I objected, because it was checking that, and you changed it so that it wasn't. Now, it's half-and-half. In one case it checks, and the other doesn't, which seems a bit off to me. So, if you're sure that there's not really any relation between front and the seed type (and the more I think about it, the more it seems like there wouldn't be), then isSeedable shouldn't be checking the type of front at all.

[jpf91]

OK, I removed the check again. It doesn't really make sense and user code should always check isRandomRNG(Type, frontType) and isSeedable(SeedType).

[jmdavis]

Merged.