Prepare for removal of FillArrays._copy_oftype

[devmotion]

JuliaArrays/FillArrays.jl#273 plans to remove the internal function FillArrays._copy_oftype since it is not used anymore in FillArrays. It seems ArrayLayouts imports this internal function and hence will be broken if it is not available anymore. This PR fixes the problem by defining the helper function in ArrayLayouts itself if FillArrays._copy_oftype is not available.

Alternatively, a slightly simpler approach would be to unconditionally define _copy_oftype in ArrayLayouts - at the cost of an additional method if FillArrays._copy_oftype exists.

When this PR is merged, we can rerun the downstream tests in JuliaArrays/FillArrays.jl#273 to ensure that the ArrayLayouts master is not broken by the FillArrays PR.

cc: @jishnub

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.03 🎉

Comparison is base (a7786a0) 86.49% compared to head (d5c959e) 86.52%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #148      +/-   ##
==========================================
+ Coverage   86.49%   86.52%   +0.03%     
==========================================
  Files          10       10              
  Lines        1740     1744       +4     
==========================================
+ Hits         1505     1509       +4     
  Misses        235      235              

Impacted Files	Coverage Δ
src/ArrayLayouts.jl	82.32% <100.00%> (+0.19%)	⬆️
src/mul.jl	82.07% <100.00%> (+0.17%)	⬆️

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[jishnub]

I think it's ok to add it here without referring to FillArrays. Bugfixes may be simpler that way

[devmotion]

OK, I simplified the PR 🙂

[jishnub]

Could you add some tests for these?

[devmotion]

Which tests do you have in mind? I assumed it would already be tested at least indirectly since the function is already used by ArrayLayouts (and hence not "new").

[jishnub]

That's possible, and in that case, no further tests are necessary. I just wasn't certain if both the methods were tested. But then the coverage data suggests that it's indeed being tested.

[devmotion]

Honestly, it's not clear to me if it was actually tested (indirectly) properly. So I just copied the tests from FillArrays (which are a bit special and FillArrays-specific) and some more general tests.

[devmotion]

Good to merge and release @jishnub (I guess it's obvious that I'd like to get the FillArrays PR merged to fix the Turing issues 😛)?

[jishnub]

Looks good to me

[jishnub]

Could you bump the patch version?

[devmotion]

Done 👍