Skip to content

DiffEqBase: add a global_error group to DEVerbosity - #4981

Open
jClugstor wants to merge 1 commit into
SciML:masterfrom
jClugstor:diffeqbase_global_error_verbosity
Open

jClugstor wants to merge 1 commit into
SciML:masterfrom
jClugstor:diffeqbase_global_error_verbosity

Conversation

@jClugstor

Copy link
Copy Markdown
Member

This PR adds five message toggles to DEVerbosity for GlobalDiffEq, in a new group global_error. GlobalDiffEq uses them in the events and DAE support of GlobalErrorEstimation and GlobalAdjoint, in a PR that follows this one. With the toggles, the warnings of GlobalDiffEq follow the verbose keyword of solve, as the warnings of the solvers do.

Toggle Message
global_error_event_mismatch The events or the terminations of the forward solve and of the true solution differ.
global_error_companion_order GlobalErrorEstimation solves its companion with a solver of order 3 or lower.
global_error_final_solve Under global error control, the final forward solve does not repeat the accepted solve.
global_error_ad_fallback GlobalAdjoint cannot use AD for the model or a callback, and uses finite differences or a numerical VJP.
global_error_grazing_event An event of GlobalAdjoint is grazing, so its saltation matrix has no time-shift term.

Presets:

  • None: all toggles are silent.
  • Minimal: the event mismatch, the final solve and the grazing event warn. The companion order and the AD fallback are silent, because they are about cost and accuracy only.
  • Standard, Detailed and All: all toggles warn.

DEVerbosity(global_error = SciMLLogging.Silent()) sets the five toggles together, and a toggle keyword overrides the group.

The PR only adds fields, so the version changes from 7.21.5 to 7.22.0. A new test in test/verbose_inference.jl checks the group keyword, the None and Standard presets, and the type stability of _process_verbose_param.

🤖 Generated with Claude Code

Add five message toggles for GlobalDiffEq in a new group global_error:
global_error_event_mismatch, global_error_companion_order,
global_error_final_solve, global_error_ad_fallback and
global_error_grazing_event. None silences them; Minimal keeps the event
mismatch, the final solve and the grazing event; Standard, Detailed and
All warn for all of them. Add a test for the group keyword and the presets.

The PR only adds fields, so the version changes from 7.21.5 to 7.22.0.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ChrisRackauckas

Copy link
Copy Markdown
Member

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

PR: #4981

Reviewed head: 09a91f2
Raw SHA-256: c4feb5dc5a7f64647ad08b4784c8a01e0ad84d854e737db01982c5c364c9b6a0
Source SHA-256: c1b316d0d133844fe2d35d651b8f29da5803ae36b85eaa6524293b85997c99bf
Source: standard-review-v1 (harness-log)
Reviewer session: 01a1d257-7f28-76f2-9585-d36a42a0ac56
Slot: 1
Confidence: unknown
Own risk fixed before author rating/discussion: yes

Risk assessment

  • Risk: low (L8)
  • Verdict: MERGE
  • Reviewed head: 09a91f2

Review report

VERDICT: MERGE
Round: 1
Head: 09a91f2
Slot: 1
CI: CI is not green (regress:sublibrary-ci / lib/OrdinaryDiffEqTaylorSeries / Julia 1 / Tests - Core (Julia 1); sublibrary-ci / lib/OrdinaryDiffEqTaylorSeries [QA] / Julia 1 / Tests - QA (Julia 1); sublibrary-ci / lib/OrdinaryDiffEqTaylorSeriesImplicit / Julia 1 / Tests - Core (Julia 1); sublibrary-ci / lib/OrdinaryDiffEqTaylorSeriesImplicit [QA] / Julia 1 / Tests - QA (Julia 1))
Session: 01a1d257-7f28-76f2-9585-d36a42a0ac56

Risk

low (L8: optional verbosity keywords on an existing constructor; every pre-existing preset field remains equal to origin/master in executed comparisons).

Blocking findings

None. This is code-review advice; the supplied failing CI remains a separate merge gate.

  • Verdict: MERGE
  • Round: 1
  • Blocking priorities: none
  • Deferred P2 issues: none

Non-blocking findings

None.

What I ran

AI reviewer: codex-cli 0.164.0, model gpt-6.1-sol; local session 01a1d257-7f28-76f2-9585-d36a42a0ac56 (no session URL).

  • Inspected git diff origin/master...HEAD, affected sources/tests, docs inclusion, and the risk criteria. No root AGENTS.md or CLAUDE.md exists in this checkout. git diff --check origin/master...HEAD passed. The reviewed checkout stayed clean and at the requested SHA.
  • Ran Julia 1.13.2 in an exact HEAD archive under scratch, using DiffEqBase's project and inherited layered JULIA_DEPOT_PATH; Pkg.instantiate() resolved SciMLLogging 2.1.0. Used two Julia threads, four precompile tasks, a 6 GiB systemd scope, and a 3,600-second test timeout after resource admission. No committed Julia manifest or interpreter lock exists.
  • Included lib/DiffEqBase/test/verbose_inference.jl: inference tests 2/2, new group tests 4/4 passed; the @inferred checks completed without error.
  • Ran independent scratch probes: 3,427/3,427 passed. Exercised all five presets, all five new fields, four message levels, group/individual precedence, unrelated-field isolation, unchanged original fields against the base implementation, inference, default isolation, actual warning emission, and lazy silence. The initial run had five reviewer-probe failures because its matcher omitted SciMLLogging's toggle prefix. Matching the supplied message payload while retaining warning-level and emission-count checks corrected that probe assumption; no package code changed.
  • Ran ENV["GROUP"] = "Core"; Pkg.test(; julia_args = ["--startup-file=no", "--check-bounds=yes"]) in the staged DiffEqBase project. Exit 0; output: Testing DiffEqBase tests passed. Core reported 292.54 seconds, including Verbose Inference | 6 6.
  • Ran uv run --locked python --version: Python 3.15.0, with the warning that --locked has no effect outside a project. There is no uv.lock or Python project here, so this establishes no locked Python environment.

What I did not verify

  • The four supplied failing TaylorSeries/TaylorSeriesImplicit CI jobs, including whether they reproduce on the default branch; CI is not established green by this review.
  • QA, documentation build, the full monorepo/downstream suites, or other supported Julia versions.
  • End-to-end GlobalErrorEstimation/GlobalAdjoint consumers; the changed logging configuration and actual message emission were exercised directly.
  • PR body, comments, author risk rating, or any previous review. No GitHub writes, commits, pushes, or checkout edits were made.

AUTHOR_GRADE: {"task": "feature", "score": 5, "difficulty": 1, "rounds": 1, "author_harness": "unknown", "author_model": "unknown", "note": "Focused additive verbosity controls with documentation and meaningful tests; affected tests, DiffEqBase Core, and 3427 independent probe assertions passed. Existing preset fields matched the base implementation. Supplied CI failures were not investigated."}

Head link: 09a91f2


🤖 Posted by an AI agent — harness: codex-cli 0.164.0 · model: gpt-6.1-sol
Conversation: local session 01a1d257-7f28-76f2-9585-d36a42a0ac56 (no session URL)

@ChrisRackauckas

Copy link
Copy Markdown
Member

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

The MERGE review recorded at head 09a91f24 cannot count toward a merge because CI at that head is red on jobs that pass on master. Seven PR-specific failures (compared against master with rack ci compare):

  • sublibrary-ci / lib/OrdinaryDiffEqTaylorSeries Core and QA
  • sublibrary-ci / lib/OrdinaryDiffEqTaylorSeriesImplicit Core and QA
  • downgrade-sublibraries / test (lib/OrdinaryDiffEqTaylorSeries) and (lib/OrdinaryDiffEqTaylorSeriesImplicit)
  • Documentation / Build and Deploy Documentation

Not diagnosed here. A similar TaylorSeries downgrade failure on another PR was traced to TaylorDiff 0.3.5 and fixed on master in #4994, so merging current master into this branch may clear it. The head is from 2026-11-13, so a rebase or merge also makes the existing review stale and a fresh review will be needed on the new head.

Five other failures and two downstream jobs (ProbNumDiffEq, SciMLSensitivity Core2) match master.


🤖 Posted by an AI agent — harness: Claude Code 2.1.325 · model: claude-sonnet-5-5
Conversation: local session 3878b54e-28f2-4a01-a176-a8f5468d1997 (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