Add ExplicitImports.jl checks to QA via SciMLTesting run_qa

[ChrisRackauckas-Claude]

Note

Please ignore this PR until it has been reviewed by @ChrisRackauckas.

What

Wires all six ExplicitImports.jl checks into the QA test group through SciMLTesting's run_qa(...; ExplicitImports = ExplicitImports, explicit_imports = true, ei_kwargs = ...):

• no_implicit_imports
• no_stale_explicit_imports
• all_explicit_imports_via_owners
• all_qualified_accesses_via_owners
• all_qualified_accesses_are_public
• all_explicit_imports_are_public

Violations and how they were resolved (FIX > DECLARE-PUBLIC > IGNORE)

print_explicit_imports(FunctionProperties; report_non_public = true) surfaced exactly two violations, both internal Core accesses inherent to hasbranching, which is a compiler-introspection utility:

Name	Where	Check	Resolution
Core.GotoIfNot	using Core: GotoIfNot (src line 3)	all_explicit_imports_are_public	minimal documented ignore
Core.Typeof	Core.Typeof.(x) (src line 47)	all_qualified_accesses_are_public	minimal documented ignore

Neither can be cleanly fixed:

• Core.GotoIfNot is the IR node type that hasbranching scans code_typed output for. It is non-public in Core and has no public alias.
• Core.Typeof builds the dispatch-signature tuple type passed to code_typed. It is not interchangeable with the public typeof: for a type-valued argument x = Int, Core.Typeof(Int) is Type{Int64} (the dispatch signature) while typeof(Int) is DataType. The equivalent Base.typesof is itself non-public, so swapping would just relocate the violation.

Both are therefore added to a minimal, documented per-check ei_kwargs ignore-list (the source line itself is unchanged, so no behavior change). FunctionProperties has no genuinely-public-but-unexported names of its own, so no public/export declarations were needed.

Also in this PR

Folds Aqua and JET into the same run_qa call and drops the two stale @test_broken false placeholders (the #54 workarounds):

• Aqua now passes with deps_compat enabled — the root Project.toml already carries [compat] for every test extra, so the deps_compat = false workaround is obsolete.
• JET typo-mode passes — the Cassette.m is not defined failure was eliminated when Fix Julia 1.12: replace Cassette with code_typed-based branch detection #55 replaced Cassette with code_typed.

Leaving @test_broken false in place would now error as an unexpected pass, so removing the placeholders and running the real checks is the correct fix.

Test deps

• test/qa/Project.toml: add ExplicitImports (compat 1.15), bump SciMLTesting floor to 1.4 (the release whose run_qa supports ExplicitImports).
• Root Project.toml: bump SciMLTesting compat floor to 1.4.

Verification (local, Julia 1.12)

GROUP=QA julia --project=. -e 'using Pkg; withenv("GROUP"=>"QA") do; Pkg.test("FunctionProperties"); end':

Test Summary: | Pass  Total   Time
QA/qa.jl      |   18     18  41.6s
     Testing FunctionProperties tests passed

The six ExplicitImports checks individually (with the configured ignores):

Test Summary:                       | Pass  Total  Time
ExplicitImports (6 checks)          |    6      6  2.8s
  no_implicit_imports               |    1      1
  no_stale_explicit_imports         |    1      1
  all_explicit_imports_via_owners   |    1      1
  all_qualified_accesses_via_owners |    1      1
  all_qualified_accesses_are_public |    1      1
  all_explicit_imports_are_public   |    1      1

Runic-clean.

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Converted the QA group to the SciMLTesting 1.5.0 run_qa minimal form (1.5.0 is now released and registered).

qa.jl collapses to:

using SciMLTesting, FunctionProperties, JET, Test
run_qa(FunctionProperties; explicit_imports = true,
    ei_kwargs = (; all_explicit_imports_are_public = (; ignore = (:GotoIfNot,)),
                   all_qualified_accesses_are_public = (; ignore = (:Typeof,))))

Aqua and ExplicitImports are SciMLTesting 1.5 direct deps and are auto-detected, so they are no longer threaded as module args. explicit_imports = true is kept (it defaults to false / opt-in in 1.5). jet auto-detects from using JET. The ei_kwargs ignore-lists for the two Core accesses inherent to hasbranching are retained.

Deps: dropped ExplicitImports from test/qa/Project.toml (now transitive via SciMLTesting); bumped SciMLTesting floor to 1.5 (root + qa env).

Aqua kept as a direct dep on purpose. Dropping Aqua from test/qa/Project.toml makes Aqua's Method ambiguity check error with Package Aqua not found in current path: that check spawns a fresh subprocess via Base.load_path_setup_code() and does using Aqua against the active QA project's load path, where only direct deps resolve. Every other Aqua check runs in-process via SciMLTesting's own Aqua binding and is unaffected, and the six ExplicitImports checks run in-process too — so only the ambiguity subprocess needs Aqua listed here. A load-path fix for that subprocess belongs upstream in SciMLTesting; until then the direct-dep entry stays (documented with a comment in the file).

Verified locally against released SciMLTesting 1.5.0 (GROUP=QA Pkg.test), QA group 18/18 on both Julia 1.10 (lts) and 1.12:

• Aqua 11/11 (incl. Method ambiguity)
• JET 1/1
• ExplicitImports 6/6 (no_implicit_imports, no_stale_explicit_imports, all_explicit_imports_via_owners, all_qualified_accesses_via_owners, all_qualified_accesses_are_public, all_explicit_imports_are_public)

Same checks and coverage as the previous commit. Branch was already current with main (rebase was a no-op). Please ignore until reviewed by @ChrisRackauckas.