Remove dead code from Random

[mgold]

Previously, seeds carried around two unused functions split and range. Every seed carried around the same next function. By removing these, we slightly decrease the memory footprint of a Seed and reduce the runtime size by, I haven't measured, but a few hundred bytes.

You could go further by placing the State directly in the Seed value constructor without the use of a record, but I'm trying to be minimally invasive.

[rtfeldman]

Assuming this works, any reason not to do type Seed = Seed State and save a record?

[rtfeldman]

Assuming this works, any reason not to do type Seed = Seed State and save a record?

[mgold]

As mentioned above, none except being minimally invasive.

Evan has expressed interest in replacing the RNG with PCG so in that larger PR I can be more assertive.

[mgold]

As mentioned above, none except being minimally invasive.

Evan has expressed interest in replacing the RNG with PCG so in that larger PR I can be more assertive.

[rtfeldman]

Gotcha. 👍 Makes sense to me!

[rtfeldman]

Gotcha. 👍 Makes sense to me!

[evancz]

Can you link to the implementation of the alternate approach that is "more random"?

[evancz]

Can you link to the implementation of the alternate approach that is "more random"?

[mgold]

type Int64
  = Int64 Int Int

type Seed
  = Seed Int64 Int64

Then this code. Also, these 64-bit arithmetic helpers.

If you're not planning to allow splittable seeds, you can eliminate one Int64 from the seed and make it a private constant. But I strongly suggest you include independent seeds because they're quite useful (for the multithreaded tests runner Richard is writing among other things). Also, although one frequently talks about splitting seeds themselves, I've found the generator of a seed to be a better abstraction. Trust me, I've spent a lot of time thinking about this so you don't have to :)

[mgold]

type Int64
  = Int64 Int Int

type Seed
  = Seed Int64 Int64

Then this code. Also, these 64-bit arithmetic helpers.

If you're not planning to allow splittable seeds, you can eliminate one Int64 from the seed and make it a private constant. But I strongly suggest you include independent seeds because they're quite useful (for the multithreaded tests runner Richard is writing among other things). Also, although one frequently talks about splitting seeds themselves, I've found the generator of a seed to be a better abstraction. Trust me, I've spent a lot of time thinking about this so you don't have to :)

[mgold]

That (about independent seeds) being said, it's possible to incorporate the better RNG now and add independent seeds later. Don't let that prevent you from adding the RNG to a patch release.

[mgold]

That (about independent seeds) being said, it's possible to incorporate the better RNG now and add independent seeds later. Don't let that prevent you from adding the RNG to a patch release.