Implement DataValueInterfaces definitions

[quinnj]

No description provided.

[quinnj]

Bump

[davidanthoff]

I'm completely swamped through the end of next week and won't be able to engage on this until then. I'll try to follow up then, but if I forget, feel free to bump again :)

[quinnj]

Bump again

[davidanthoff]

I'll have some time this week to look at this!

[quinnj]

@davidanthoff, I've updated this PR to point to https://github.com/JuliaData/AbstractDataValues.jl (happy to move the package to Queryverse if you'd like; I'm not sure it really matters where it lives, as long as we both have admin access).

One thing I realized from what we talked about: having DataValues overload Base.nonmissingtype doesn't quite achieve the same purpose as nondatavaluetype because we're really looking for the Union{T, Missing} equivalent of DataValue{T} (and vice-versa for datavaluetype), which conflicts with the purpose of Base.nonmissingtype, which is to get the T from Union{T, Missing} (so the equivalent DataValue definition would be Base.nonmissingtype(::Type{DataValue{T}}) where {T} = T). I also left unwrap in for now as I haven't tested moving things over to using convert. I'll give that a shot and report back if we can remove it, but wanted to ping you on the initial setup here first.

[davidanthoff]

Cool!

I think it might make slightly more sense to host the package in queryverse? That way the two DataValue packages would be close together. But I also don't feel strongly about it. Oh, the one (really not very important) benefit would be that it would show up at https://www.queryverse.org/status/.

Should we maybe call the package DataValuesInterface.jl? Abstract to me implies a type, which we don't really have in that package?

re nondatavaluetype, yep, I see your point. I have two unrelated thoughts on that: a) should it be called missingtype, to be symmetric with datavaluetype? Maybe that is slightly more specific, by telling a user what they will get? b) I wonder whether this could still be folded into something existing, though... Maybe something like convert(Union{nonmissingtype(typeof(x)), Missing}, typeof(x)? But that is probably an abuse of convert... And maybe we should also not let the perfect be the enemy of the working in this case...

Re unwrap, yes, lets see whether you can get rid of that somehow. There is also x[] that unwraps, not sure whether that might also work?

One other random thought: the whole exercise does start to feel a little like defining an interface for different missing types... Probably not where we want to go?

[quinnj]

Sounds good. Could you create the repo in Queryverse org then and then add me as (at least) a collaborator? Then I can do the initial setup and update the PR here.

[quinnj]

Ok, I've pushed all the code for new DataValueInterfaces.jl: https://github.com/queryverse/DataValueInterfaces.jl. Tables.jl PR is here. I'll add some tests here.

[davidanthoff]

Looks good. I probably have to trigger the registration of DataValueInterfaces? Let me know when you are ready with all the rest, and then I'll trigger the registration, and then we can merge this here.

[quinnj]

I think we're ready; let's trigger DataValueInterfaces.jl registration and I'll do a little more stress-testing on Tables.jl in the mean time.

[davidanthoff]

JuliaRegistries/General#1630

[davidanthoff]

@quinnj I just restarted tests, and I think they are missing a using DataValueInterfaces?

[quinnj]

Alright, let's see how this goes (everything passes on my machine)

[codecov-io]

Codecov Report

Merging #57 into master will decrease coverage by 0.09%.
The diff coverage is 66.66%.

@@           Coverage Diff            @@
##           master     #57     +/-   ##
========================================
- Coverage    74.7%   74.6%   -0.1%     
========================================
  Files          13      13             
  Lines         506     512      +6     
========================================
+ Hits          378     382      +4     
- Misses        128     130      +2

Impacted Files	Coverage Δ
src/scalar/core.jl	92.59% <66.66%> (-1.21%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b6b0735...8b927d5. Read the comment docs.

[quinnj]

Ok, tests look good here. Once we merge and tag a release here, I can re-run tests here and merge/tag there. Getting close!

[davidanthoff]

JuliaRegistries/General#1708