Allow version 0.19 of DataStructures

[andreasnoack]

Might be good to set up CompatHelper to simplify this although there is only a single dependency.

Might also be better to test on pre instead of nightly.

[DilumAluthge]

Looks like CI isn't using DataStructures 0.19.

Example CI log: https://github.com/JuliaCollections/SortingAlgorithms.jl/actions/runs/16706927424/job/47286361683?pr=101

⌅ [864edb3b] DataStructures v0.18.22

[andreasnoack]

Okay. Then I'll set up CompatHelper. That will make it easier to resolve.

[LilithHafner]

CompatHelper won't fix this. The problem is that we load StatsBase in testing and StatsBase does not declare compatibility with .19. This is a chicken and egg problem with JuliaStats/StatsBase.jl#959. The solution is to trick Julia & Pkg into disregarding StatsBase (and other test deps') declared incompatibility with DataStructures .19. @DilumAluthge, this seems like something that would come up fairly frequently. Is there a built in mechanism to do this?

[DilumAluthge]

I think we have two different problems here.

Problem 1: Detecting this situation

In this PR, CI looks green, but didn't use the desired version of DataStructures.

In julia-runtest, we have special logic to detect CompatHelper PRs. If a CompatHelper PR is detected, then julia-runtest will force the latest version of dependencies (and will throw an error if that's not possible). So the underlying incompatibility still exists, but at least we get a visible red X in CI, and thus it's easier for us to notice that this problem exists.

This logic (in julia-runtest) only applies to CompatHelper PRs. It doesn't apply to human PRs, which is why julia-runtest doesn't run that logic in this PR, and thus why we have a deceptive green CI checkmark on this PR.

Problem 2: Chicken/egg problem between this package and StatsBase

I don't know of a built-in mechanism for solving this problem. It may be worth opening an issue on Pkg.jl.

One manual workaround for us to use here might be:

1. Modify the CI script in this repo to manually Pkg.develop() the CompatHelper branch of StatsBase (the branch for CompatHelper: bump compat for DataStructures to 0.19, (keep existing compat) JuliaStats/StatsBase.jl#959).
2. Get CI passing here with the latest 0.19 of DataStructures, and merge PR here.
3. Tag new release here.
4. Wait for new release of this report to be merged into General and propagated to Pkg servers
5. Rebase CompatHelper: bump compat for DataStructures to 0.19, (keep existing compat) JuliaStats/StatsBase.jl#959 (or just close/reopen the PR, if it is already up-to-date with master).
6. Now CI on CompatHelper: bump compat for DataStructures to 0.19, (keep existing compat) JuliaStats/StatsBase.jl#959 should be using 0.19 of DataStructures.
7. Merge CompatHelper: bump compat for DataStructures to 0.19, (keep existing compat) JuliaStats/StatsBase.jl#959
8. Tag new release of StatsBase
9. Wait for new release of this report to be merged into General and propagated to Pkg servers
10. REVERT step 1 (the Pkg.develop() in this repo).

[LilithHafner]

CompatHelper is also good because that will give watchers a notification whenever a breaking release of DataStructures is cut.

[LilithHafner]

I'd squeeze step 10 into step 2 between "Get CI passing here with the latest 0.19 of DataStructures" and "merge PR here."

[devmotion]

Can we just remove StatsBase from the list of test dependencies? Circular dependencies are bad in general IMO.

[LilithHafner]

In this case, yes. But in general, that's not always viable. It's quite reasonable to want to use tooling packages when testing and it's also reasonable for tooling to have deps.