More robust test/examples.jl

[tkf]

This is a PR onto #200.

I think the test is failing because Base.current_project() doesn't work nicely with --project:

arakaki@amdci2 ~/.j/d/KernelAbstractions (vc/split)> pwd
/home/arakaki/.julia/dev/KernelAbstractions
arakaki@amdci2 ~/.j/d/KernelAbstractions (vc/split)> julia --startup-file=no --project=test -e 'display(Base.current_project())'
"/home/arakaki/.julia/dev/KernelAbstractions/Project.toml"⏎
arakaki@amdci2 ~/.j/d/KernelAbstractions (vc/split)> julia --startup-file=no --project=test -e 'display(Base.load_path())'
3-element Array{String,1}:
 "/home/arakaki/.julia/dev/KernelAbstractions/test/Project.toml"
 "/home/arakaki/.julia/environments/v1.5/Project.toml"
 "/home/arakaki/opt/julia/julia-1.5.2/share/julia/stdlib/v1.5"⏎

A quick workaround may be to use Base.load_path_setup_code() to propagate load path etc to the child processes.

[vchuravy]

bors try

[bors]

try

Build failed:

• CI-julia-1-6-nightly (1.6-nightly, ubuntu-latest, x64)

[vchuravy]

Fascinatingly now it is only partly broken

[tkf]

Maybe the failure in Windows is due to the incompatibility between powershell and bash?

https://github.com/JuliaGPU/KernelAbstractions.jl/runs/1882919388?check_suite_focus=true#step:6:3

I just added the patch that is supposed to use bash everywhere.

I'm not sure why it'd fail on ubuntu, though https://github.com/JuliaGPU/KernelAbstractions.jl/runs/1882919107#step:9:128

I can't repro it locally

[vchuravy]

bors try

[bors]

try

Build failed:

• buildkite/kernelabstractions-dot-jl/julia-1-dot-6-nightly

[vchuravy]

bors help

[vchuravy]

bors delegate=@tkf

[DilumAluthge]

I think it's

bors delegate=tkf

[bors]

✌️ tkf can now approve this pull request. To approve and merge a pull request, simply reply with bors r+. More detailed instructions are available here.

[tkf]

Thanks, Dilum!

[tkf]

bors try

[bors]

try

Build failed:

• CI-julia-1-6-nightly (1.6-nightly, ubuntu-latest, x64)

[tkf]

bors try

[bors]

try

Build failed:

• buildkite/kernelabstractions-dot-jl/julia-1-dot-6-nightly

[tkf]

bors try

[tkf]

bors try

[bors]

try

Already running a review

[tkf]

bors try

[bors]

try

Already running a review

[DilumAluthge]

Why all this business with the load path? Why not just do Pkg.test("KernelAbstractions")?

[vchuravy]

Ha! If it were just that simple, Pkg.test doesn't like the mono-repo setup we are going for.

[tkf]

So, it looks like this is a Julia 1.6-specific problem JuliaLang/julia#39631

(I think this is orthogonal to the issue that Pkg.test is not enough for this mono-repo setting.)

[vchuravy]

right you need to add @stdlib

[tkf]

I believe the behavior of 1.6 is a bug. You shouldn't need anything in the top-level dependencies for transitive dependencies. If it's in the manifest, it should work.

[tkf]

OK, github action worked in my repo tkf#1 (esp https://github.com/tkf/KernelAbstractions.jl/runs/1883606478) so I think this is good to go.

I'm not merging via bors since I don't know if it supports merging to a non-default branch.

[bors]

try

Timed out.

[vchuravy]

Thanks Taka!

[DilumAluthge]

Ha! If it were just that simple, Pkg.test doesn't like the mono-repo setup we are going for.

I'm not sure I follow. You should be able to do the following:

julia test.jl

Where the contents of test.jl are:

import Pkg

# KernelAbstractions
Pkg.develop(Pkg.PackageSpec(path = pwd()))

# CUDAKernels
Pkg.develop(Pkg.PackageSpec(path = joinpath(pwd(), "lib", "CUDAKernals")))

Pkg.test(["KernelAbstractions", "CUDAKernels"])

Does that not work? If it doesn't work, we should patch Pkg to make it work.