[ChrisRackauckas]

🚧 UNREVIEWED — awaiting review by @ChrisRackauckas. Filed by an AI agent running as @ChrisRackauckas; Chris has not seen this.
Harness: Devin CLI 3000.11.3 (diagnosis and fix), Claude Code 2.1.285 (PR) · Model: fusion-claude-opus-5-5-high-sidekick-swe-2-medium
Conversation: local Claude session 3c311569-dda6-4687-bc9f-65febb212c33 (fleet master)

This fixes a failure on the default branch's CI. The diagnosis and the evidence below come from the investigating agent's report; "confirmed by running" and "read in logs" are kept apart there.

Repo state: main at 578e7f2 ("Bump versions for release (#97)"). On main, Tests, Downgrade, Downstream and format-check all passed for that commit. Two workflows failed on main:

Workflow	Run	Cause	Category
Documentation (push)	https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38483093460	gh-pages push race with the PR #97 preview deploy	FLAKY (CI race)
format-pr (schedule)	https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38489959826	workflow calls julia without installing it; the job ran on a self-hosted runner that has no Julia	repo CI config bug, triggered by runner assignment → FIX-READY

---

1. Documentation: git push to gh-pages rejected (FLAKY)

Error (job "build / Build and Deploy Documentation", https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38483093460/job/115590134982):

[gh-pages 2a798c4] build based on 578e7f2
To https://github.com/SciML/FastAlmostBandedMatrices.jl.git
 ! [rejected]        HEAD -> gh-pages (fetch first)
error: failed to push some refs to 'https://github.com/SciML/FastAlmostBandedMatrices.jl.git'
ERROR: LoadError: failed process: ... git push -q upstream HEAD:gh-pages ...
in expression starting at .../docs/make.jl:26

The docs build passed (makedocs and HTML rendering succeeded). Only deploydocs failed.

Cause (read in logs and the GitHub API, not reproduced locally): two Documentation runs pushed to gh-pages at the same moment:

• The main push run 38483093460 (runner self-hosted-4vcpu-16gb-8q5rj-runner-wvlx9) and the PR Bump versions for release #97 run 38483089399 (https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38483089399, runner ...-m4tng) were created 2 s apart (23:15:14/16Z). Both waited about 2.5 h for self-hosted runners and both started at 01:52:2xZ.
• docs/make.jl uses deploydocs(; push_preview = true). The PR run logged Deploying preview build ... Deploying: ✔ for pull/97/merge = d7f24a7.
• The gh-pages history (GET /repos/SciML/FastAlmostBandedMatrices.jl/commits?sha=gh-pages) has a1bb9a2 2026-10-14T01:54:41Z build based on d7f24a7. That is the PR preview commit, landed in the same second the main run made its local commit 2a798c4. The main run's push 1 s later was a non-fast-forward, so it was rejected.
• After that, the tag run for v0.1.14 (https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38499425926) deployed 1b74a36 build based on 578e7f2 successfully. The previous main Documentation run (38326969146) also passed.

So this is a one-off race between concurrent deploys. It is not a bug in this repo or in a dependency. Impact: dev/ on gh-pages was not refreshed for 578e7f2. That commit only bumps the version, so the content is unchanged. Re-running the job fixes it.

Possible hardening, not done here because it is in the shared workflow: the reusable SciML/.github/.github/workflows/documentation.yml@v1 could serialize gh-pages deploys with a concurrency group, or retry the push. Documenter itself does not retry a rejected push.

Existing PR: none (searched open PRs for gh-pages/documentation). Fix made: none. No in-repo fix is warranted.

---

2. format-pr: julia: command not found (FIX-READY)

Error (job "build", https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38489959826/job/115612559840, step "Install JuliaFormatter and format"):

Runner name: 'self-hosted-4vcpu-16gb-8q5rj-runner-h6fkl'
##[error]Process completed with exit code 127.

Cause (read in logs): .github/workflows/FormatPR.yml declares runs-on: ubuntu-latest and runs julia -e ... with no julia-actions/setup-julia step. It only works when the job gets a GitHub-hosted ubuntu-24.04 image, which ships with Julia. For example, the passing run 38327648800 used Image: ubuntu-24.04 and Julia 1.13 from the image. When the job is assigned to an org self-hosted runner, Julia is not on PATH. This is intermittent and long-standing, not a new regression. Every failed format-pr run among the last 40 has the same error on a self-hosted runner:

• 2026-10-14 https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38489959826 (8q5rj-runner-h6fkl)
• 2026-09-27 https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/36282748236 (zfxvv-runner-28xjz)
• 2026-09-24 https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/35938418111 (d6j5t-runner-z2ktl)
• 2026-09-13 https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/34728119167 (x2k99-runner-flfvk)
• 2026-09-09 https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/34295069381 (x2k99-runner-wn5ld)
• 2026-09-05 https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/33932943849 (7smtr-runner-tl56q)

The missing setup step dates back to the commit that introduced the workflow (786de00, "Setup CI", per the PR #88 bisect).

Second problem with this workflow (confirmed by running): the repo has been formatted with Runic since PR #45, and format-check (SciML/.github/.github/workflows/runic.yml@v1) enforces it. FormatPR runs JuliaFormatter with default style, so the PR it keeps refreshing, #67 "Automatic JuliaFormatter.jl run", rewrites the code into a style that fails Runic:

$ # main (578e7f2)
$ julia +1.12 --project=@runic -m Runic --check benchmark docs src test
main exit=0

$ # PR #67 head (JuliaFormatter output), extracted to scratch/pr67
$ julia +1.12 --project=@runic -m Runic --check --diff benchmark docs src test
pr67 exit=1
-struct DisjointRange{T<:Integer,R1<:AbstractUnitRange{T},R2<:AbstractUnitRange{T}} <:
-       AbstractVector{T}
+struct DisjointRange{T <: Integer, R1 <: AbstractUnitRange{T}, R2 <: AbstractUnitRange{T}} <:
+    AbstractVector{T}
...
-        return @inbounds d.r2[i-n1]
+        return @inbounds d.r2[i - n1]

(full diff: scratch/runic_pr67.txt). When the workflow does pass, its output is a PR that would turn format-check red if merged. Runic suggestions on PRs already come from RunicSuggestions.yml.

Existing PRs:

• Set up Julia in formatter workflow #88 "Set up Julia in formatter workflow" by @ChrisRackauckas-Claude. It added julia-actions/setup-julia@v3 to FormatPR.yml. @ChrisRackauckas closed it unmerged on 2026-09-07, with no comment. It would have fixed the exit-127 error, but it would have kept the JuliaFormatter-vs-Runic conflict.
• Automatic JuliaFormatter.jl run #67 "Automatic JuliaFormatter.jl run" (github-actions bot, open since 2026-06-16, CI action_required). This is the workflow's own output, not a fix. It should be closed.
• Precedent elsewhere in the org: Delete FormatPR.yml workflow SteadyStateDiffEq.jl#111 "Delete FormatPR.yml workflow" (merged).

Test before/after: this is a scheduled CI workflow, so no Julia test can cover it, and I could not run the GitHub Actions job locally. Before the fix, the workflow exists and fails whenever it lands on a self-hosted runner (6 runs listed above). When it does succeed, it produces Runic-violating output (Runic --check exit 1 on PR #67, shown above). After the fix, the workflow is gone, so the scheduled run that failed no longer exists. Formatting is still enforced by format-check (Runic --check exit 0 on main, shown above).

If you'd rather keep an automatic formatting PR, the alternative is to replace FormatPR.yml with a Runic-based job that has a setup-julia step. I did not do this because it is a design choice for you to make.

---

Checks not run

• I could not exercise the GitHub Actions workflows locally, so there is no actionlint or act run.
• I did not run the package test suite: the only change is deleting a CI workflow, and Tests are green on main.

Links

• Failed Documentation run (main): https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38483093460
• Concurrent PR Bump versions for release #97 Documentation run that won the gh-pages race: https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38483089399
• Successful v0.1.14 tag docs deploy: https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38499425926
• Failed format-pr run (main): https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38489959826
• Last passing format-pr run (GitHub-hosted runner): https://github.com/SciML/FastAlmostBandedMatrices.jl/actions/runs/38327648800
• PR Set up Julia in formatter workflow #88, setup-julia in FormatPR (closed unmerged): Set up Julia in formatter workflow #88
• PR Automatic JuliaFormatter.jl run #67, JuliaFormatter auto-PR that fails Runic (open): Automatic JuliaFormatter.jl run #67
• PR Switch from JuliaFormatter to Runic.jl for code formatting #45, switch to Runic: Switch from JuliaFormatter to Runic.jl for code formatting #45
• PR Bump versions for release #97, release bump whose preview deploy raced: Bump versions for release #97
• Org precedent, SteadyStateDiffEq.jl FormatPR deletion: Delete FormatPR.yml workflow SteadyStateDiffEq.jl#111

---

🤖 Posted by an AI agent — harness: Devin CLI 3000.11.3 (fix), Claude Code 2.1.285 (PR) · model: fusion-claude-opus-5-5-high-sidekick-swe-2-medium / claude-opus-5-5
Conversation: local Claude session 3c311569-dda6-4687-bc9f-65febb212c33 (fleet master)

🤖 Generated with Claude Code