Platform independent pseudo random sequences

[Alpar Juttner]

LEMON's random generator is based on the Mersenne Twister algoritm and it is optimized to 32 and 64 bit architecture. This is fine, except when someone want to obtain reproducible results (for a given initial seed) on each platform. The attached patch provides the following solution to this.

The original Random class becomes a template and goes to the _random_bits namespace, then three typedefs are defined:

• Random is exactly the same as the class Random was before,
• Random32 is the 32 bit version (on all platform) and
• Random64 is the 64 bit version (on all platform).

This solution is fully backward compatible and the only drawback I can see is the documentation - one has to visit the template version to see the detailed API.

What do you think, is the documentation is acceptable in this form? If not, what would be a better solution?

[Peter Kovacs]

I do like this changeset, we should merge it to the main branch.

I also added a patch [026829fdb6d4] with javadoc improvements (it is on the top of the proposed patch).

[Alpar Juttner]

I cannot apply 602-026829fdb6d4.patch​ with --exact, mercurial says this patch is corrupt.

[Peter Kovacs]

Replying to alpar:

I cannot apply 602-026829fdb6d4.patch​ with --exact, mercurial says this patch is corrupt.

Strange. I can reproduce the issue, but I do not understand its cause. It seems that if I export a changeset after it was modified (e.g. hg commit --amend), then it cannot be imported with --exact. But if I import it without --exact, it works and a new hash ID is added to the commit. If I export this new version, then it can be imported with --exact. I do not understand why this happens. (I use Mercurial 4.3.1 on Ubuntu 17.10)

I recreated the patch as [04f57dad1b07] and tested it, it should work correctly.

[Alpar Juttner]

Yes, now it works. Thanks.

[Alpar Juttner]

Changesets [db1d342a1087] and [04f57dad1b07] are pushed to the main branch.