[C++] Implement "random" nullary function that generates uniform random between 0 and 1

[asfimport]

This is similar to PostgreSQL's random()

https://www.postgresql.org/docs/8.2/functions-math.html

Reporter: Wes McKinney / @wesm
Assignee: Alex Suhan / @asuhan

Related issues:

• [C++] Random real generator is slow on Arm64 Linux when built with clang (is related to)
• [C++] Return a random sample of rows from a query (is depended upon by)

PRs and other links:

• GitHub Pull Request #11864

_{Note: This issue was originally created as ARROW-12404. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
@edponce

[asfimport]

Eduardo Ponce / @edponce:
I will work on this, but want to be clear with the requirement. Arrow has non-nullary uniform random number generators that generate N numbers which are defined in the testing module (.e.g, random_real()). The nullary random() will be included in the random namespace and make use of the GenerateOptions?

[asfimport]

Antoine Pitrou / @pitrou:
No, this shouldn't use any functionality defined in testing, as it is not built by default. Also, we won't necessarily want to use the same random number generator or initialization parameters.

(see also ARROW-10797 for a slightly related topic)

[asfimport]

yibocai#1:
If we are to implement from scratch a method other than std::uniform_real_distribution , it's better to take into account issue ARROW-12533.

[asfimport]

Eduardo Ponce / @edponce:
Hi @cyb70289,

I am working on this issue and noticed that you submitted #10283 for ARROW-12533. Can I duplicate the RNGs for the compute layer?

[asfimport]

yibocai#1:
Sure, @edponce

[asfimport]

Eduardo Ponce / @edponce:
Thanks! I know that code duplication is a "code smell" but compute and testing layers are independent of each other. As I understand Arrow does not (yet) has a well-defined common submodule to supply utility functions across layers, and this might be further limited if APIs of same functions vary across layers.

[asfimport]

Antoine Pitrou / @pitrou:
Utility code goes into arrow/util/ ;)

[asfimport]

Antoine Pitrou / @pitrou:
That said, I'll reiterate that a RNG for testing needn't have the same properties as a RNG exposed as a public API. We need to be a bit more careful about the public API RNG:

• needs to have good-quality statistical output
• needs to have consistent, predictable performance

[asfimport]

Eduardo Ponce / @edponce:
Thanks for feedback @pitrou. I will definitely be careful with the public RNG. Will circle back on this when I get a stable implementation.

[asfimport]

Antoine Pitrou / @pitrou:
Issue resolved by pull request 11864
#11864