Require SciMLTesting 2.4

[ChrisRackauckas-Claude]

Ignore this PR until it has been reviewed by @ChrisRackauckas.

Summary

• require SciMLTesting 2.4 in the root test environment
• supersede closed Require SciMLTesting 1.2 for grouped test harness #4804 following the maintainer direction to drop all pre-2.4 releases
• leave source code and public API unchanged

The existing SymbolicIndexingInterface group exercises the grouped test harness and dependency resolution affected by this compat entry, so no separate source-level test is needed.

Local validation

• Julia 1.10.11: GROUP=SymbolicIndexingInterface julia +1.10 --project -e 'using Pkg; Pkg.test()' — 1,964/1,964 passed with SciMLTesting 2.4.0
• Julia 1.12.6: the same official group — 1,964/1,964 passed with SciMLTesting 2.4.0
• repository-wide Runic 1.7 --check --diff . — passed
• git diff --check — passed

No test was skipped, silenced, or loosened.

[ChrisRackauckas-Claude]

Validation scratchpad for 00fa42992e4b5d97f378741bf5e129605a25b3ce on base 165213e965b4dceea08b18526dbaa82fc3e60e41:

• Julia 1.10.11 with SciMLTesting 2.4.0, using the repository's official GROUP=SymbolicIndexingInterface entrypoint: 1,964/1,964 passed in 25m01.0s. Raw-log SHA-256: a3171a80f8d730fe9107180380fa6b83754bfd21705c1b6a66d9e6bfcf41d423.
• Julia 1.12.6 with SciMLTesting 2.4.0, using the same official group: 1,964/1,964 passed in 28m33.3s. Raw-log SHA-256: 1dd60e89737963ae523a299d5d910e0cb1e9a8d0140ca4f70cc2f8024c21a4f4.
• Repository-wide Runic 1.7 --check --diff . passed. Raw-log SHA-256: 1111d1680399a66b11f01107f77ef64d5f288a21b67cdb2f14c6239254ed3f91.
• git diff --check passed.
• The generated ignored Manifest.toml was removed; the feature worktree is one commit/one file and clean.

This supersedes closed #4804 and implements the maintainer's explicit “Dropping all pre 2.4” direction. No test was skipped, silenced, or loosened.

[ChrisRackauckas-Claude]

Checked after the two-hour no-poll boundary at 2026-07-21 21:36:09 EDT.
Connector and git readback both reported draft PR #4807 at exact head
00fa42992e4b5d97f378741bf5e129605a25b3ce, based on current master
165213e965b4dceea08b18526dbaa82fc3e60e41. The local branch and remote branch
were identical, and HEAD..origin/master was empty.

Successful exact-head workflows included root Downgrade, Sublibrary CI,
documentation, both format checks, spell check, and benchmarks. The old shared
sublibrary downgrade workflow also reported SciCompDSL success, but that result
predates SciML/.github #119 and is
not true old-style-extra floor evidence.

Failed checks

• ReleaseTest Catalyst/Modeling/1 job 88791277427 reproduced current-master
  job 88779812274: both failed Symbolic Stoichiometry with 20 passes and one
  error. This is the already classified Catalyst clean-base Modeling failure,
  not a Require SciMLTesting 2.4 #4807 regression.
• Downgrade Sublibraries ModelingToolkitBase job 88791302406 reproduced
  current-master job 88783044520 exactly: 7,203 passes, 14 failures, 21
  errors, and four broken tests. Both have the same 18
  idxs_i == length(idxs) + 1 assertion errors and the same
  Float64(::ForwardDiff.Dual) boundary-value errors. It is assigned as a
  separate clean-base investigation and is not being conflated with Require SciMLTesting 2.4 #4807.
• Root QA job 88791325832 and current-master QA job 88784389630 both ended
  while the test step still reported in_progress; all later steps remained
  pending, and GitHub returned no job log. This is runner/service-loss evidence,
  not a Julia test failure.
• Root Optimization failed on all three Julia lanes on both current master and
  Require SciMLTesting 2.4 #4807. For the Julia 1.10 lane, current master reported 125 passes, six
  failures, and one broken test; Require SciMLTesting 2.4 #4807 reported 123 passes, eight failures, and
  one broken test. The source diff is only the root SciMLTesting compat line,
  and normalized installed package/version sets are identical between the two
  jobs (the only diff is the checkout path prefix). Five failure locations are
  common. Different runs additionally fail different numerical solver
  assertions at lines 42 versus 52/76/77, with different computed solutions,
  establishing an existing unstable clean-base test rather than a changed
  runtime dependency graph. This still requires its own clean-base
  investigation; it is not described as passing.
• IntegrationTest ModelingToolkitStandardLibrary/Core job 88791278395
  reproduced the clean-base so₊u(t) error at test/utils.jl:41. Draft
  ModelingToolkitStandardLibrary PR #491
  contains the locally verified fix.
• The two Catalyst Modeling failures in IntegrationTest are current-base
  failures already classified during
  Catalyst PR #1519 work;
  Require SciMLTesting 2.4 #4807 changes no Catalyst or ModelingToolkit runtime source.

The candidate's intended floor has direct local evidence: the official
SymbolicIndexingInterface group passed 1,964/1,964 on Julia 1.10.11 and
1.12.6 with SciMLTesting 2.4.0, and the corresponding hosted Julia 1, LTS, and
prerelease jobs all passed. This classification does not call the whole PR CI
green; it separates the one-line floor change from current-master failures that
must be repaired independently.