CI never seems to complete with Documenter 1.0

[DanielVandH]

In SciML/FiniteVolumeMethod.jl#50, I'm testing the deployment with Documenter v1 vs. Documenter v0.27. In the v0.27 case here, everything runs fine. In the v1 case here, it just never finishes. I don't think this is a fluke either, since I also noticed it a lot in a previous PR (this one, but I made pr50 to show this issue since there's too many commits in pr49).

Is there anything particular about the make.jl file that would be causing this in v1 versus v0.27? The make.jl is here https://github.com/DanielVandH/FiniteVolumeMethod.jl/blob/doc/docs/make.jl

[mortenpi]

I would guess that somehow the at-example blocks are getting stuck, which might be due to some IO blocking etc. I am not sure why 0.27 and 1.0 would be different here though -- I don't think we had any significant changes to the at-block evaluation.

What you could do is set JULIA_DEBUG=Documenter and re-run the CI. You should get more debug output, maybe pointing to where it's getting stuck.

[DanielVandH]

I turned on the debug, but nothing seems to happen:

https://github.com/DanielVandH/FiniteVolumeMethod.jl/actions/runs/6211290382/job/16860337346?pr=50

It's weird cause it works fine locally... Any other debugging ideas?

[mortenpi]

No good ideas. I would remove the julia-actions/julia-docdeploy action, since it does weird stuff, and see if anything changes. It might be eating up the debug output.

It does look like it gets stuck somewhere. So one option would be to try to make the process print a stack trace every now and then. Starting up this thread before makedocs might work for that:

Threads.@spawn begin
    @info "async SIGUSR1 loop started"
    while true
        sleep(5*60) # every 5 minutes, adjust as needed
        run(`kill -SIGUSR1 $(getpid())`)
    end
    @info "async SIGUSR1 loop terminated"
end

[DanielVandH]

Thanks @mortenpi. I'll try it out and see.

[DanielVandH]

I tried removing docdeploy and the CI runs fine. Putting docdeploy back in and putting that thread you've given into make.jl, I get the output here:

https://github.com/DanielVandH/FiniteVolumeMethod.jl/actions/runs/6213368725/job/16864357366?pr=50

I don't really know what I'm reading from these logs. Does anything stand out to you?

[mortenpi]

I tried removing docdeploy and the CI runs fine.

Then I would move the issue there and use the standard setup in the meanwhile. It does not look like a Documenter bug. Nothing specifically stands out though. It looks like it's stuck in this function somehow:

Documenter.jl/src/documents.jl

Lines 905 to 913 in c3fc63b

	function addpage!(doc::Document, src::AbstractString, dst::AbstractString, wd::AbstractString)
	page = Page(src, dst, wd)
	# page's identifier is the path relative to the `doc.user.source` directory
	name = pagekey(doc, page)
	# This check is here to make sure that the new function matches the old
	# (correct) implementation.
	@assert name == normpath(relpath(src, doc.user.source))
	doc.blueprint.pages[name] = page
	end

My best guess would be that it's the loading of GitHubActions.jl from a different environment brings in conflicting dependencies, and that just creates some weird behavior. But I am not entirely convinced of that either.

[DanielVandH]

I've posted it over on the docdeploy repo. Thanks for the help @mortenpi. I'll close this since it's not related to Documenter.