Zero dimensional BlockIndex

[mtfishman]

Fixes #430.

I bumped the minor package version since this is mildly breaking, though I think it would be strange if code was relying on the previous behavior and this PR could be considered a bug fix.

To-do:

• Add tests.

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 4 lines in your changes missing coverage. Please review.

Project coverage is 0.00%. Comparing base (1e5feaa) to head (88f4db7).
Report is 2 commits behind head on master.

Files with missing lines	Patch %	Lines
src/blockindices.jl	0.00%	4 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (1e5feaa) and HEAD (88f4db7). Click for more details.

HEAD has 7 uploads less than BASE

Flag	BASE (1e5feaa)	HEAD (88f4db7)
	9	2

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #431       +/-   ##
==========================================
- Coverage   93.67%   0.00%   -93.68%     
==========================================
  Files          18      18               
  Lines        1643    1630       -13     
==========================================
- Hits         1539       0     -1539     
- Misses        104    1630     +1526     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[mtfishman]

You can see that there were some tests that were explicitly relying on the previous definition that Block()[] == Block(), but the current definition Block()[] == BlockIndex() follows from the property that a[Block()[]] == a[Block()][] so I think that the previous definition must have been a mistake.

[mtfishman]

I see there is a downstream test failure in LazyBandedMatrices.jl here: https://github.com/JuliaLinearAlgebra/LazyBandedMatrices.jl/blob/v0.10.4/test/test_blockconcat.jl#L348 which is calling: https://github.com/JuliaLinearAlgebra/LazyBandedMatrices.jl/blob/v0.10.4/src/blockconcat.jl#L489.

The root cause of that seems to be that Block is not being treated as a scalar in broadcasting, it seems like maybe it was working accidentally before this PR.

[mtfishman]

I see now that Block(4)[] was defined to return Block(4) before because it was matching the behavior of Number, i.e. 4[] === 4, but that seems like a strange choice given that indexing into Block returns BlockIndex (for good reason) in all other cases. I'm testing out changing the broadcasting behavior of Block to define it as a scalar in broadcasting to see if that fixes the LazyBandedMatrices.jl test failure but also allows the BlockArrays.jl tests to pass.

[dlfivefifty]

Do we know why the LazyArrays.jl test is failing?

[mtfishman]

Do we know why the LazyArrays.jl test is failing?

I don't know, but it seems to be failing in the same way in other PRs. It looks like an issue with how the LazyArraysBlockArraysExt package extension is configured. I see that BlockArrays is listed as both a dependency and a weak dependency of LazyArrays: https://github.com/JuliaArrays/LazyArrays.jl/blob/v2.2.1/Project.toml#L8, that doesn't seem right.

[dlfivefifty]

Having it listed as both was necessary and standard to support Julia before extensions so that shouldn’t be the issue…

[mtfishman]

Having it listed as both was necessary and standard to support Julia before extensions so that shouldn’t be the issue…

I see. I see a few issues related to package extension loading in Julia 1.11 like JuliaLang/julia#55939 and JuliaLang/julia#56204 but I'm not sure if they are related.

But also I see that LazyArrays.jl requires Julia 1.10 anyway so it seems like they can just be weakdeps now and the code logic https://github.com/JuliaArrays/LazyArrays.jl/blob/v2.2.1/src/LazyArrays.jl#L80-L85 can be removed. EDIT: I'm trying that out in JuliaArrays/LazyArrays.jl#347.

[mtfishman]

I reran the LazyArrays.jl integration test now that JuliaArrays/LazyArrays.jl#347 is merged and registered but there is still an issue.

Maybe the issue is being triggered because BlockBandedMatrix.jl has BlockArrays.jl as a dependency, and the package system is getting confused about that. Again maybe related to JuliaLang/julia#55939.

[jishnub]

Many of these issues will be resolved in Julia v1.11.2. Could you check on that branch? Or perhaps we may wait for the release, which should be soon.

[dlfivefifty]

I thought the lazyarrays extensions were rewritten to avoid this problem…

[mtfishman]

I thought the lazyarrays extensions were rewritten to avoid this problem…

I'm having trouble reproducing the issue locally so it is tough for me to test, JuliaArrays/LazyArrays.jl#347 was my best guess for a quick fix but it must be another issue. As @jishnub said, probably it is best to wait for Julia v1.11.2 to be released and rerun the tests against that version.

[jishnub]

You may test against Julia nightly meanwhile, as any feature being backported to v1.11 would be present on nightly

[mtfishman]

Ok, I reran the downstream tests against nightly, and the LazyArrays.jl tests don't hit that same installation issue (one test fails for some unrelated issue, but at least the tests run). So that's a good indication that we can wait for the Julia v1.11.2 release and that will hopefully resolve this issue. I'll revert back to testing against the latest stable release, and I guess we can wait for that release to test again and then merge this.

[mtfishman]

I've tried running the downstream tests against a few Julia versions at this point, here is the summary:

• Julia v1.11.0: All tests pass.
• Julia v1.11.1: LazyArrays.jl tests fail when LazyArraysBlockArraysExt is being loaded with the error "ERROR: LoadError: ArgumentError: Package LazyArraysBlockArraysExt does not have BlockArrays in its dependencies", all others pass.
• Julia v1.11.2 pre-release: LazyArrays.jl tests fail with the same error as in Julia v1.11.1, all others pass. I tested it by setting 1.11-nightly in the workflow.
• Julia nightly: The last time I checked LazyArrays.jl tests ran properly (so the original issue plaguing Julia v1.11.1 and v1.11.2 is fixed) but failed for another independent reason.

I haven't been able to reproduce this locally so I'm at a loss for how to proceed. @jishnub any suggestions? It seems like there is an issue in Julia v1.11.1 that is still in the v1.11.2 pre-release related to package extension loading. I would like to raise an issue about this to the Julia (or Pkg) Github repository but if I can't reproduce it locally I can't create a minimal reproducible example...

[dlfivefifty]

I think just file an issue with a link to the CI failures showing the regression. It definitely seems like a Julia problem.

[dlfivefifty]

If its independent of this PR then we can merge this

[mtfishman]

I think just file an issue with a link to the CI failures showing the regression. It definitely seems like a Julia problem.

Sounds good. I agree the issue is independent of this PR, let's merge this and I can start a new PR that just demonstrates the LazyArrays.jl loading issue and I'll make an issue on the Julia repo about it.