Stale deps test misses `Statistics` in `BlockBandedMatrices.jl`

[jishnub]

In https://github.com/JuliaLinearAlgebra/BlockBandedMatrices.jl/actions/runs/5672874507/job/15373054076, the Aqua ("project quality") test passed, but the package has Statistics as a stale dependency that it doesn't load. The Aqua command that it uses is

Aqua.test_all(BlockBandedMatrices, ambiguities=false, piracy=false,
        # only test formatting on VERSION >= v1.7
        # https://github.com/JuliaTesting/Aqua.jl/issues/105#issuecomment-1551405866
        project_toml_formatting = VERSION >= v"1.9")

so stale deps should be tested for. Oddly, this leads to a test failure in a downstream test in FillArrays.jl (I'm unsure if this has just been fixed).

In case it matters, one of the example files (in the examples directory) loads Statistics

[lgoettgens]

If I am not mistaken, Statistics is a stdlib, right?
If yes, Aqua does not guarantee to detect it not being used (see https://juliatesting.github.io/Aqua.jl/stable/#Aqua.test_stale_deps).

Nevertheless, it is weird that it behaves differently in the two different test cases. I will dig into this.

[lgoettgens]

Ok, so I have an explanation now.
In BlockBandedMatrices.jl before JuliaLinearAlgebra/BlockBandedMatrices.jl#164 (so e.g. v0.12.3), Statistics.jl is not considered a stale dependency, as it gets loaded by the command using BlockBandedMatrices. You are right, it does not get loaded as a direct dependency, but as a dependency of FillArrays.jl. Aqua cannot distinguish the two cases.

In JuliaArrays/FillArrays.jl#286, Statistics.jl gets added as a weakdep of FillArrays.jl, and is thus no longer considered a dependency for julia versions 1.9+.

Newer Julia versions will ignore dependencies in [deps] that are also in [weakdeps].
(see https://pkgdocs.julialang.org/v1/creating-packages/#Transition-from-normal-dependency-to-extension)

Using this version of FillArrays.jl, neither BlockBandedMatrices.jl nor any of its dependencies loads Statistics.jl. Thus it is considered stale and makes the Aqua test fail.

So overall, there is no problem here, and everything is working as expected.

[jishnub]

Aqua cannot distinguish the two cases

This should perhaps be clarified in the documentation, although it does warn about stdlibs.

Thanks for digging into this!

[lgoettgens]

Aqua cannot distinguish the two cases

This should perhaps be clarified in the documentation, although it does warn about stdlibs.

This does not only happen for stdlibs. I added some clarification in #164.