Revisit Julia 1.6 support

[ablaom]

Currently we have julia="1.6 in the Project.toml. If we are explicitly using Xoshiro RNG's , then we can only support >= 1.7. On the other hand, CI is only testing 1.8 and the latest version (now 1.9).

I think for now we should:

• [ ] Bump compat to julia="1.7" and that we add 1.7 to the test matrix.

• [ ] Then, as a second step, we can swap out Xoshiro for MersenneTwister in special case VERSION <= v"1.7" to extend support to the Long Term Support release Julia 1.6. I suspect 1.6 will be the LTS for a while yet. This likely requires some special casing for tests, so may be a little work??

But we could also do both at once.

[EssamWisam]

Alrighty. Working on a small PR after adding some implementation notes in the docs and other improvements. Once done will bump Julia compat to 1.7 and add it to the test matrix.

[EssamWisam]

@ablaom, Okay I am a little confused. Why is the second step needed if we'll bump compat to 1.7? No one with Julia VERSION <= v"1.7" can install Imbalance.jl in this case and Xoshiro (which I do use explictily in algorithms when an integer is given, tests and examples) is available.

[ablaom]

I was only suggesting the bump to 1.7 as a temporary measure.

Longer term, I thought it would be nice to support Julia 1.6 (revert the compat to julia="1.6") because it is the Julia Long Term Support release. (Some research institutions, eg defense-related research, are limited to how often they can update their software stacks, for security reasons. So I think it is a good idea to support the older platform, until Julia itself stops supporting it.) But we can't do this without some kind of fix for the fact that Xoshiro was not introduced until 1.7.

One possiblity is to explicitly introduce version-specific code to switch out Xoshiro for MersenneTwister if VERSION <= V"1.7", and adjust tests if necessary.

Another breaking solution is to always use MersenneTwister and adjust tests if necessary. (BTW, I don't think a breaking release at this early stage is a big deal.)

I thought we might use use Random.default_rng() but that is pre-seeded. Some variations on that idea are not working for me, because copy is not implemented for MersenneTwister in Julia 1.6

Maybe the best solution is to just forget about supporting 1.6 and do something if it is requested. I see SciML has already dumped support for Julia <= 1.9, probably because they like the pkg extension feature.

[ablaom]

Okay, I would like to push a little harder for 1.6 support. See for example this discussion. Also, MLJ supports 1.6 and so the planned integration is currently failing CI.

[EssamWisam]

Okay, I would like to push a little harder for 1.6 support. See for example this discussion. Also, MLJ supports 1.6 and so the planned integration is currently failing CI.

Sorry. I should have said that I have absolutely agreed with your prior message (makes lots of sense to support LTS) and that I scheduled myself to finish working on it today. Other issues besides the RNG popped up and currently dealing with them. In fact, RNG wasn't really hard to solve.

[EssamWisam]

@ablaom Julia 1.6 support is here now. This is the new RNG dosctring:

    "RNG" => """
    - `rng::Union{AbstractRNG, Integer}=default_rng()`: Either an `AbstractRNG` object or an `Integer` 
        seed to be used with `Xoshiro` if the Julia `VERSION` supports it. Otherwise, uses MersenneTwister`.
    """,

[ablaom]

Perfect. Then I guess you can tag a new patch release and close this issue.