[ocots]

Written by Claude (Claude Code) at the request of ocots.

Closes #405. Refs #310, #327.

What

tools/ci/scope.py computes environments × problems from the files a pull request changes, and the scope labels only add to it. The rules are read from the repository, not written as a list of paths:

• the environments are the directories of test/suite/ with a Project.toml, without the opt-in ones (OPT_IN of test/runtests.jl);
• ext/<M>Models/<p>.jl gives M and comparison for p; another file of ext/<M>Models/ gives M and comparison for every problem;
• ext/<E>.jl gives the environments whose Project.toml holds every trigger of E ([extensions] of Project.toml);
• a file of ext/MetaData/ or ext/Descriptions/ named after a problem gives every environment for that problem (now including ext/MetaData/references/<p>.toml, which ran everything before);
• test/suite/<env>/** gives env;
• test/common/<f>.jl gives the environments whose tests name <f>, directly or through test/common/; every environment if test/runtests.jl names it (options.jl, run_environment.jl and what it includes);
• test/cron/**, test/robustness/**, tools/**, docs/** and Markdown outside ext/: core only;
• anything else, and every fallback of [Tests] Test only the problems a pull request changes #327 (removed or renamed file, unreadable or truncated list, release, push): everything.

Labels: test <M> backend and test comparison add an environment (with the problems of the automatic scope, every problem if it has none: one product, see the comment on #405); test all problems widens to every problem; test everything (new label) runs everything; test convergence unchanged. The job summary has one row per file: its environments, problems and why.

test/README.md (section Continuous integration, with the table of the rules) and the comments of CI.yml are updated. Nothing changes on the Julia side.

Done when (#405)

• scope.py computes environments × problems with the rules; unit tests for every rule and fallback, on a fixture repository, shown to fail on broken rules (below).
• Labels add; test everything runs everything (the label itself is to be created on the repository: see below).
• The job summary shows, for each file, its environments and problems.
• Shown on recent pull requests (below), instead of on this one (it changes CI.yml, so it runs everything, as the rules say).
• test/README.md and the comments of CI.yml up to date.

Tests

python3 -m unittest discover -s tools/ci/test -p 'test_*.py'
Ran 47 tests in 4.310s
OK

Three rules broken on purpose (not committed), then restored:

== mutation: no comparison for a modeling
FAIL: test_all_problems_label, test_convergence_label, test_environment_label_adds,
      test_problem_of_a_modeling, test_shared_file_of_a_modeling, test_union, test_narrowed
FAILED (failures=7)
== mutation: no transitive include
FAIL: test_shared_test_file, test_shared_test_file_of_the_dispatcher
FAILED (failures=2)
== mutation: labels restrict
FAIL: test_environment_label_adds, test_environment_label_never_restricts
FAILED (failures=2)

The new script on the files of recent pull requests (status from the API):

PR	Files	Before	Now
#400 (scalarized equivalence)	2	everything	comparison
#402 (schemes by features)	6	everything	everything (a removed workflow)
#396 (#371)	11	everything	everything (a removed test file)
#395 (#377)	96	everything	everything (src/)
#380 (references)	40	everything	everything (Project.toml)

And on single files of this repository: ext/JuMPModels/beam.jl → JuMP comparison problem=beam; ext/JuMPModels/schemes/build.jl → JuMP comparison; ext/OptimalControlModels.jl → OptimalControl OptimalControlScalarized comparison; test/common/methods.jl → JuMP OptimalControl OptimalControlScalarized; test/common/status.jl → everything; docs/make.jl → core.

To do before merging

• Create the label test everything on the repository (not done: an action on the repository, waiting for ocots).

No runner label: the CI tooling tests run in the scope job without it, and nothing changes on the Julia side.

Main checkout git status --short: empty before and after.

Conventions: ct-conventions v0.10.0, ct-agents v0.4.0.

🤖 Generated with Claude Code