Fix doctest

[ViralBShah]

Fix doctest seen in:

https://buildkite.com/julialang/julia-master/builds/44616#0194f191-ee57-4a81-ab55-0192700d1df9

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.87%. Comparing base (ed35a37) to head (9d0a260).
Report is 23 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1199   +/-   ##
=======================================
  Coverage   91.87%   91.87%           
=======================================
  Files          34       34           
  Lines       15366    15366           
=======================================
  Hits        14118    14118           
  Misses       1248     1248           

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

[KristofferC]

This doesn't look right, or? Maybe

Suggested change

	3×3 LowerTriangular{Float64, {Matrix{Float64}}:
	3×3 LowerTriangular{Float64, Matrix{Float64}}:

[dkarrasch]

There must be a mismatch between used code and tests in the failing doctest. This has been changed deliberately lately IIRC.

[KristofferC]

My guess is that the docs don't run with the custom sysimage and are in fact testing the bundled LinearAlgebra. Could that be correct?

[ViralBShah]

Can someone look at https://buildkite.com/julialang/julia-master/builds/44616#0194f191-ee57-4a81-ab55-0192700d1df9/823-1641?

I don't see why Adjoint is in the output for LowerTriangular in the CI, both before and after:

2025-02-10 16:13:20 EST | │   diff =
-- | --
  | 2025-02-10 16:13:20 EST | │    3×3 LowerTriangular{Float64, Adjoint{Matrix{Float64}}}:
  | 2025-02-10 16:13:20 EST | │      Adjoint{Float64, Matrix{Float64}}}:
  | 2025-02-10 16:13:20 EST | │      2.0   ⋅    ⋅
  | 2025-02-10 16:13:20 EST | │      6.0  1.0   ⋅
  | 2025-02-10 16:13:20 EST | │     -8.0  5.0  3.0

None of this got caught in the CI in this repo - and at least in my local build on master it seems to do what I expect, which is what the PR does (modulo manual editing and messing up - and wondering why CI passed):

[dkarrasch]

#1186

[ViralBShah]

I still don't understand how these tests pass here but not on julia master.

[ViralBShah]

Ok - the issue in this repo seems to be that the CI/docs workflow only generates the docs but does not run the doctests.

[dkarrasch]

I quickly ran the doctest in my REPL manually, and it does show the Adjoint, so this PR is certainly going in the wrong direction. Should we close it and open another one on fixing the whole sysimg-stdlib-doctest-whatnot issue?

[ViralBShah]

This PR is certainly incorrect. We should do the whole fix the sysimg doctest thing. But there is a separate point about whether the Adjoint is a good idea.

I suppose fix the sysimg thing and this doctest so that we can bump LinearAlgebra on julia master once again. Afterwards, if we can decide if we want to revert #1186.

[dkarrasch]

I'm closing this. @KristofferC spotted the typo in the doctest (which was truly failing) and fixed it.