Release Distributions.jl v1.0

[non-Jedi]

I'm eager to have the guarantees of semver for packages I use as dependencies. What needs to happen before this package could be considered 1.0 and released as stable? From the outside it seems like it might already be ready for that, but I don't have a good view on any breaking changes that may still need to be made or essential functionality that's missing.

[simonbyrne]

#823 is definitely a big breaking change.

[matbesancon]

@non-Jedi thanks for this issue.

• As mentioned, Release StatsBase.jl v1.0 StatsBase.jl#493 is a big one. Once done, we have to migrate all StatsBase functions
• WIP: Add keyword constructors #823 made us realize the tests are in a bad state, we should remove JSON-based tests and replace them with generated Julia files, will be much faster
• Improved documentation for sampler and distribution interfaces. In the current state, it's very hard to know what is required and what is optional for implementing new distributions. The doc should be improved for that.
• if done at all, Remove UnivariateGMM #844 needs to be done before freezing
• there are some design issues with mixtures
• removing dependencies on PDMats would make the lib way more lightweight cc @andreasnoack MvNormal Matrix Types and Speed #688

Given how much work that already is on the plain stuff which would make Distributions 1.0-ready, I tend to think #823 is optional there, if not present, it should be a 2.0 milestone

[datnamer]

What's needed for GPU support?

[nalimilan]

* [](https://github.com/JuliaStats/Distributions.jl/issues/688)

Given how much work that already is on the plain stuff which would make Distributions 1.0-ready, I tend to think #823 is optional there, if not present, it should be a 2.0 milestone

It would be weird to tag 1.0 with a large PR that everybody is looking forward. At least, deprecations could be made so that the PR can be merged without breaking anything in 1.x.

OTC, refactoring tests doesn't affect the API, so I don't see what it should be on the 1.0 roadmap. Same for documentation.

[matbesancon]

@datnamer

What's needed for GPU support?

I am not sure what you mean. In doubt, I'd say generic containers everywhere instead of hardcoded Float64

[matbesancon]

Refactoring tests is a major point blocking #823.
I was considering 1.0 not only from the semver point of view, but also as signal of readiness and robustness of the library. Documentation here is also documenting clearly what is required as interface for a distribution, otherwise any removal afterwards is breaking

[simonbyrne]

Refactoring tests is a major point blocking #823.

Not really: we could merge #823, it's just that the keyword constructors aren't really being tested.

[matbesancon]

Not really: we could merge #823, it's just that the keyword constructors aren't really being tested.

Not testing the keyword constructor would be problematic, we would merge something we are not able to test against moving forward.

Ok for #823 before 1.0 yes, even if we don't need to see 2.0 as a super-distant milestone, unlike the Julia project itself. DifferentialEquations has been moving on with major releases

[matbesancon]

@non-Jedi if you want to pick up some issues from there feel free :)
I tagged a bunch of issues as up-for-grabs

[matbesancon]

Also: @cscherrer, this one is still open and could potentially lead to breaks
#771

[cossio]

Please consider also #911, which I think could also lead to breaking changes.

[matbesancon]

It's not major, it could lead to breaking change but nothing awful, and always fixable by specifying types