Getting a package into JuliaStats

[bdeonovic]

What is involved in putting a package into JuliaStats? Who has administrative privileges?

[andreasnoack]

The main criterion for including a package in JuliaStats is the number of collaborators. If several people need commit access it can be convenient to have the repo in an organization. I assume you are asking because of Mamba.jl. Does the criterion apply there?

I think there might be some reservations to include more packages into JuliaStats because it sends a signal of collective responsibility of maintenance and with limited manpower that might be hard to fulfill.

[bdeonovic]

Brian Smith would actually like to remain primary maintainer of Mamba.jl, so I don't see Mamba.jl leeching man power away from other projects. It does seem like having the package in JuliaStats would make it much more visible to the public.

[bdeonovic]

If a package does wind up in an organization, who gets commit access? Who decides who gets commit access?

[nalimilan]

Brian Smith would actually like to remain primary maintainer of Mamba.jl, so I don't see Mamba.jl leeching man power away from other projects. It does seem like having the package in JuliaStats would make it much more visible to the public.

The risk is not really that a project would "leech man power", it's rather that nobody would really maintain it, which would go against the signal sent by hosting the project in JuliaStats. The hosting makes the most sense when several members of JuliaStats work actively on a project, and consider themselves as responsible for it.

[bdeonovic]

Brian and I (especially Brian) are very active on the project. Just look at the commits...

[johnmyleswhite]

FWIW, here's my two cents:

• I think it's fine in this specific case to include Mamba.jl in JuliaStats since it's clearly been actively maintained for a long time by Brian and you.
• In general, our biggest cultural problem is that we need people to stop writing new packages and to focus their energy maintenance work on the packages that already exist.

The second issue makes me very sympathetic with Milan's concern, even though I'm happy to pull in Mamba.jl.

I think it's pretty clear that there's already too much stuff in JuliaStats as is -- and that some of it is of pretty low quality compared with the better packages. Given how dire the situation has become, I think it's very easy to agree with Milan's feeling that working on anything other than the existing packages is effectively redirecting labor that we desperately need focused elsewhere.

But, in practice, I don't think this concern can be acted on in any meaningful way because volunteers can't be forced to do the work that the core JuliaStats developers might want them to do. The only thing we can hope to do is to gradually increase the number of people who feel like they have a stake in JuliaStats and hope that things work out.

[nalimilan]

So the deal is "promise you'll contribute to StatsBase and you'll get hosted in JuliaStats"? ;-)

I should have noted that, having no specific knowledge of the field, I have no objection to including Mamba.jl. Though I wonder what's the reason for the existence of two competing implementations of MCMC.

[mschauer]

Related: #17 .

[bdeonovic]

Thanks for all of the comments. Is there a clear answer to:

If a package does wind up in an organization, who gets commit access? Who decides who gets commit access?

edit: Another question: When a package moves into an organization does the development history move along with it (commits, issues, etc)

[papamarkou]

@nalimilan, @mschauer, as your comments and references implicate Mamba and Lora as competing MCMC implementations, I would like to kindly take a silent stance. The way my creativity is wired, it works best under peaceful conditions, so I would like to not add any other comment to a potential reflame of an older holy MCMC war. I prefer to focus on doing work, as it turns to be more constructive and positive, at least for me, with regards to the specific matter :)

[mschauer]

Haha, no, the point was just showing my own experience in finding JuliaStats strangely inaccessible. As I am interested in MCMC methods on the path space of stochastic processes (MCMC inference methods using a proposal distribution for stochastic process bridges etc.) I am fully aware about the vastness of the design space for these packages.

[papamarkou]

There is no harm in adding stats-oriented packages to JuliaStats, I think some of our co-developers are only trying to raise awareness/are reaching for more effective collaboration, which is not always straightforward to achieve :)

[simonbyrne]

+1 for me as well.

If a package does wind up in an organization, who gets commit access? Who decides who gets commit access?

The org owners (of which I am one) can set user permissions. There's no set policy: typically if you're made a few good pull requests, we're happy to give commit access. Obviously in this case, we can just give it to you off the bat.

edit: Another question: When a package moves into an organization does the development history move along with it (commits, issues, etc)

Yes.