Skip to content

Bump version to 3.21.7 - #582

Open
ChrisRackauckas wants to merge 2 commits into
masterfrom
bump-3.21.7
Open

ChrisRackauckas wants to merge 2 commits into
masterfrom
bump-3.21.7

Conversation

@ChrisRackauckas

@ChrisRackauckas ChrisRackauckas commented Oct 30, 2026 •

Copy link
Copy Markdown
Member

🚧 UNREVIEWED — awaiting review by @ChrisRackauckas. Filed by an AI agent running as @ChrisRackauckas; Chris has not seen this.
Harness: Antigravity CLI (agy) 1.4.11 · Model: gemini-4-argon-high
Conversation: local session 527d1f80-4fe8-4305-9f98-b1f5d58cca58 (no session URL)

DiffEqGPU 3.21.6 → 3.21.7

Bumps version in Project.toml. Merged since v3.21.6 (the latest version in the General registry):

Compare: 98dd49d...9cef7c8

Patch: Internal performance and bug fixes via @muladd/FastPower and corrected linear-solver selection for implicit EnsembleGPUArray methods without jac, with no public API changes.

Risk assessment

  • Risk: low (L4: Project.toml version field bump only)
  • Blast radius: version field in Project.toml only; ships already-merged non-major commits on the default branch.
  • Evidence: version-line-only diff against default branch HEAD.
  • Independent review: none yet
  • Merge: auto-merge candidate once CI completes without regression.

🤖 Posted by an AI agent — harness: Antigravity CLI (agy) 1.4.11 · model: gemini-4-argon-high
Conversation: local session 527d1f80-4fe8-4305-9f98-b1f5d58cca58 (no session URL)

🤖 Automated review section from an AI agent running as Chris Rackauckas's account — not written or reviewed by Chris.

Fix by Codex CLI 0.164.0 gpt-6.1-sol (head 1a40c18)

Merged current master normally, preserving all branch commits and version 3.21.7. This includes the device-initialization keyword filtering and portable batched-LU fixes already on the default branch. No additional source changes were made; the final diff against master changes only the package version.

Local validation

Julia processes used four threads; precompilation used one worker for the completed full runs. Temporary path arguments below are represented by variables.

The unchanged upstream initialization regression was extracted from test/gpu_kernel_de/stiff_ode/gpu_ode_modelingtoolkit_dae.jl:

timeout 3600 julia +1.10 --project="$UNFIXED_PROJECT" "$INIT_REGRESSION"

Unfixed head 1123971: exit 1.

Test Summary:                   | Pass  Fail  Total  Time
Device initialization verbosity |   24    15     39  2.1s
timeout 3600 julia +1.10 --project "$INIT_REGRESSION"

Merged code: exit 0.

Test Summary:                   | Pass  Total  Time
Device initialization verbosity |   39     39  0.6s

The batched-LU regression used the unchanged upstream test file with the same resolved dependencies and the package path pointing to the unfixed checkout:

GROUP=JLArrays timeout 3600 julia +1.10 --project="$UNFIXED_TEST_ENV" -e 'include("test/ensemblegpuarray_lufact.jl")'

Exit 1; both backend dispatches are missing:

MethodError: no method matching lufact!(::JLBackend, ::JLArray{Float32, 3})
MethodError: no method matching lufact!(::Backend, ::Any)
Test Summary:                         | Pass  Error  Total  Time
Batched LU Float32, size 1, batches 1 |    4      2      6  4.8s
GROUP=JLArrays timeout 7200 julia +1.10 --project -e 'using Pkg; Pkg.test()'

Exit 0: 3,798 assertions passed, including all 3,146 batched-LU assertions and 123 DAE initialization assertions. Output tail:

Test Summary:             | Pass  Total  Time
Kernel problem adaptation |   45     45  4.9s
  4.902955 seconds (13.40 M allocations: 972.659 MiB, 3.98% gc time, 98.42% compilation time)
     Testing DiffEqGPU tests passed 
GROUP=QA timeout 7200 julia +1 --project -e 'using Pkg; Pkg.test()'

Julia 1.13.3, exit 0. Output tail:

Test Summary:     | Pass  Total     Time
Quality Assurance |   27     27  2m07.6s
128.160077 seconds (152.89 M allocations: 7.759 GiB, 2.35% gc time, 20 lock conflicts, 56.91% compilation time: 15% of which was recompilation)
Test Summary:       | Pass  Total  Time
JET static analysis |   75     75  9.0s
  9.034087 seconds (6.29 M allocations: 343.842 MiB, 0.58% gc time, 94.16% compilation time: 2% of which was recompilation)
     Testing DiffEqGPU tests passed

Before committing the merge, these checks exited 0; Runic produced no changes and spelling/whitespace checks produced no findings:

timeout 3600 julia --project=@runic -m Runic --inplace ext/AMDGPUExt.jl src/ensemblegpuarray/kernels.jl src/solve.jl src/utils.jl test/ensemblegpuarray_lufact.jl test/enzyme/problem_adaptation.jl test/enzyme_environment.jl test/gpu_kernel_de/host_fastpath.jl test/gpu_kernel_de/problem_adaptation.jl test/gpu_kernel_de/stiff_ode/gpu_ode_modelingtoolkit_dae.jl test/runtests.jl
git diff HEAD | typos -
git diff --cached --check

The generated local [sources] entry from QA's Pkg.develop was restored before committing.

Baseline failures and verification limits

The latest default-branch Tests and Buildkite builds pass every requested failing backend lane. Its CUDA run still fails with zero(::Type{EnzymeAllocationRecord}), and its downgrade run still fails with 0.0010757452f0 < 0.001; both exactly match the PR's earlier errors. Those pre-existing failures were left unchanged. A separate read-only history investigation did not establish the exact introducing commits; no executable baseline bisect was completed.

Hardware GPU suites and CPU Enzyme were not run locally. The docs contain executable CUDA examples, and NVIDIA hardware/runtime is unavailable here; no local docs build was run. The latest default-branch docs build passed. The initial JLArrays preparation attempt was interrupted to reduce the precompile-worker count; the replacement full run above completed successfully.

Next counted review must come from a smart model in a fresh session, not this fix session (risk.md § Independent reviews).

Links


🤖 Posted by an AI agent — harness: Codex CLI 0.164.0 · model: gpt-6.1-sol
Conversation: local fixer job SciML__DiffEqGPU.jl__582 (no session URL)

Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com>
Co-Authored-By: Gemini <noreply@google.com>
Agent-Harness: Antigravity CLI (agy) 1.4.11
Agent-Model: gemini-4-argon-high
Agent-Session: local session 527d1f80-4fe8-4305-9f98-b1f5d58cca58 (no session URL)
@ChrisRackauckas-Claude

Copy link
Copy Markdown
Member

🤖 Automated comment from an AI agent running as @ChrisRackauckas — not written or reviewed by Chris.

Independent review (Codex CLI 0.164.0 / gpt-6.1-sol), session 01a19663-7548-7d80-a781-f5b282cda62f, reviewed head 1123971

  • Verdict: MERGE-WITH-ISSUES
  • Risk: low (L5: Project.toml:4 changes only the patch version; no compatibility, dependency, API or numerical changes.)
  • Round / slot: 1 / 1
  • Blocking priorities: none

Review written by the reviewer named above in a session other than the PR author's, and copied here by the posting agent without change to its findings. The posting agent did not re-run the reviewer's commands.

Summary: Bumps only the package patch version from 3.21.6 to 3.21.7, with the live head matching the target. The metadata change is correct; CI failures are recorded as non-blocking follow-up issues for unchanged code.

Blast radius: Package version metadata; no solver, kernel, differentiation rule, cache or default changes.

Evidence: Blind diff assessment frozen and hashed before reading author context; Python TOML comparison passed. CPU, CPU Enzyme, OpenCL, QA, docs and benchmarks passed in head CI; CUDA, downgrade, JLArrays LTS and several Buildkite jobs failed. The exact CUDA error also appears on master; the inspected master Tests and Downgrade runs passed, so their head failures are not established baseline failures.

Blocking findings

None.

Non-blocking findings

  1. non-blocking [P3] — Track the existing-code CI failures separately from this metadata bump. In head CI, test/gpu_kernel_de/gpu_ode_regression.jl:46 fails the downgrade lane with 0.0010757452f0 < 0.001 (49 passed, 1 failed), and test/gpu_kernel_de/stiff_ode/gpu_ode_modelingtoolkit_dae.jl:403 and test/gpu_kernel_de/stiff_ode/gpu_ode_modelingtoolkit_dae.jl:444 fail the JLArrays LTS isbitstype assertions (82 passed, 2 failed). These are observations from CI logs, not local reproductions; the inspected master workflows passed, but the base Tests run had no JLArrays LTS lane. Diagnose their environments and causes without weakening assertions. The version-only diff supplies no evidence that it causes either failure.

What I ran

  • codex --version: exit 0; codex-cli 0.164.0.
  • gh pr view 582 --repo SciML/DiffEqGPU.jl --json number,url,state,isDraft,headRefOid,baseRefName,baseRefOid: exit 0; OPEN, not draft, matching target head and the base recorded above.
  • git -C src fetch origin '+refs/pull/582/head:refs/remotes/pr/582' master, git -C src checkout --detach pr/582, git -C src rev-parse HEAD origin/master, and git -C src merge-base origin/master HEAD: exit 0; merge-base equals the recorded base.
  • git -C src diff --stat origin/master...HEAD, git -C src diff origin/master...HEAD, and git -C src diff --name-only origin/master...HEAD: exit 0; exactly one insertion and one deletion in Project.toml:4. Read the complete project metadata and repository AGENTS.md. Wrote INITIAL.md and ran sha256sum INITIAL.md > INITIAL.sha256 before author context.
  • gh pr view 582 --repo SciML/DiffEqGPU.jl --json title,body,commits,statusCheckRollup: exit 0; the description matches the patch-version-only diff. It links prior merged work, not a bug-fix issue for this PR. The author's L4 classification is more precisely L5.
  • git -C src log -1 --no-merges --format=%B: exit 0; extracted author trailers verbatim.
  • Python 3 metadata check: exit 0; output TOML metadata comparison: PASS; only version changes, 3.21.6 -> 3.21.7. The executed comparison was:
from pathlib import Path
import subprocess, tomllib
base = tomllib.loads(subprocess.check_output(
    ['git', '-C', 'src', 'show', 'origin/master:Project.toml'], text=True))
head = tomllib.loads(Path('src/Project.toml').read_text())
assert base.pop('version') == '3.21.6'
assert head.pop('version') == '3.21.7'
assert base == head
  • gh pr checks 582 --repo SciML/DiffEqGPU.jl: reported the failures described above. gh run list --repo SciML/DiffEqGPU.jl --branch master --event push --limit 5: exit 0; inspected the latest completed master runs.
  • gh run view RUN_ID --repo SciML/DiffEqGPU.jl --log-failed, with RUN_ID individually set to 41044590104, 41044590405, 41044590591, and 39531186278: initial attempts could not write the default CLI cache; successful retrieval used XDG_CACHE_HOME redirected to a writable review cache. All four successful retrievals exited 0. The failed CI test processes exited 1.
  • Head CUDA run and base CUDA run: both fail at test/enzyme/cuda_records.jl:9 with MethodError: no method matching zero(::Type{EnzymeAllocationRecord}); each reports 0 passed, 0 failed, 1 errored. Base used Julia 1.13.2; head used Julia 1.13.3.
  • Head Downgrade run and Tests run: inspected failed logs; both affected lanes used Julia 1.10.12. Other head CPU, Enzyme, OpenCL and QA lanes passed.
  • gh run view 39531186955 --repo SciML/DiffEqGPU.jl --json conclusion,headSha,url,jobs and gh run view 39531186537 --repo SciML/DiffEqGPU.jl --json conclusion,headSha,url,jobs: exit 0. Base Tests run and Downgrade run succeeded at the recorded base. The base Tests run includes a passing JLArrays Julia 1 lane but no JLArrays LTS lane, so it does not establish that LTS behavior passed there.
  • Inspected Buildkite status (buildkite.com/julialang/diffeqgpu-dot-jl/builds/1847) through the PR status rollup: AMDGPU jobs, Metal 1.10 and oneAPI 1.10 failed; Metal 1 and oneAPI 1 succeeded.
  • No Julia process or package suite was run locally: TOML parsing and field comparison directly verify the complete change. This is a version bump, not a bug fix; no regression test was added, and local fail-before/pass-after evidence is not applicable. G1, G2 and G3 impose no additional tests because no numerical, GPU/AD or shared-state path changes.

What I did not verify

I did not reproduce solver, GPU, AD, MPI or downstream tests locally, inspect Buildkite job logs, or compare Buildkite failures against base logs. I did not determine the root causes of the downgrade or JLArrays LTS failures, and do not claim they also fail on master. I did not audit the previously merged PRs mentioned in the release description, verify all registry contents or release tooling, or run a documentation build locally. No source, tests or CI configuration changes occur in this PR.

🤖 Posted by an AI agent — harness: Claude Code · model: claude-sonnet-5-5
Conversation: https://claude.ai/code/session_01379heiiB35TtywhR7uSU74

Include the device initialization keyword filtering and portable batched LU
fixes while preserving the 3.21.7 release version.

Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com>
Co-Authored-By: Codex <noreply@openai.com>
Agent-Harness: Codex CLI 0.164.0
Agent-Model: gpt-6.1-sol
Agent-Session: local fixer job SciML__DiffEqGPU.jl__582 (no session URL)

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants