[vjaganat90]

What

• The REST bbbc payloads (bbbc_download_wfb.json, single_node_bbbc_download.json) run vjaganat90/bbbc-download-plugin:0.1.1-dev1.
• Dispatched Lint And Test and Run Workflows runs are titled by the commit subject, plus a paired mm-workflows or image-workflows branch when it is not main.
• 50 tests and the "PyTest Benchmark Harness" step are deleted. Each pinned how the code or the CI is built, not what a user sees. No src/ change.
• The weekly benchmark lane goes red when a case raises or its corpus preparation fails, after the full table is written. No timing fails it. The || true, the continue-on-errors and the dispatch step that failed on purpose to show the lane stays green are gone.
• test_realtime.py's cache helper writes each file in one step. The watcher thread could scan a log between its creation and its text, count a change no test made, and do an extra final run, which made test_a_failed_analysis_is_a_line intermittently report 2 runs.

Why

• REST bbbc: the old image's 2022 CA bundle cannot verify data.broadinstitute.org's chain, so both REST bbbc runs fail with CERTIFICATE_VERIFY_FAILED. This is the REST half of fix: pull the rebuilt tool images from vjaganat90 image-workflows#28.
• Run titles: push and cross-repo dispatches passed the whole commit message, and the title also listed every repo and ref, so the run list was paragraphs. The run's own branch is already shown by GitHub. The "Name what this run tests" step still writes all three checked-out commits to the summary.
• Tests: every deleted test fails on a rename, a refactor or a new enum member while users see no change. A failure there costs a fix and catches nothing. The groups and their reasons follow.

Deleted tests

• Removed names and module layout. The API simply has none of these names, and reaching for one is Python's own AttributeError/ImportError.
  • test_python_api_workflow.py: string indexing on ports, write_wic as the only writer, the top-level module list, the legacy module path, the string-keyed workflow and step methods.
  • test_python_api_tool_builder.py:
    • the old module name, no from_input on outputs, identity comparison of spec classes;
    • two of Python's own TypeErrors (positional inputs and outputs, a single FieldSpec);
    • that validation goes through cwltool in-process, which is how it runs, not what it reports.
• Release and CI config read back as text. test_release.py (7) asserted the release-please config and the YAML of the publish jobs.
• Tripwires on a single version. Sophios writes one CWL version and has one language version. Adding a second is a deliberate change that has to touch these tables anyway.
  • test_cwl_version.py: v1.2 declared, every supported version admitted.
  • test_lang_version.py: exactly one version.
  • test_inline_flags.py: the per-version table.
  • test_reference_compatibility.py: dispatch over versions.
• Inventories of the tests themselves.
  • test_cookbook.py: every construct has a cookbook pair. The pairs themselves are still tested.
  • test_manufactured_documents.py: the hand-kept list of sites no driver reaches. The checks on reached sites stay.
• Mechanism pins. The behaviour these protect is tested through parse, compile and emit.
  • test_lower.py: __post_init__ invariants, no mutable container in the graph, how often a name is printed, the output_step helper.
  • test_resolve.py: phases typed against each other, test-setup side effects.
  • test_emit.py: the compiler emits through its graph.
  • test_lang_schema.py: surface declarations, an undeclared field raising, schema objects not shared.
  • test_namespacing.py: read-only and frozen tables, membership sets computed, the exported schema not shared.
  • test_lang_parser.py: Diagnostics slicing.
  • test_lang_render.py: a literal keeping its source text.
  • test_inline_flags.py: an identity return.
  • test_diagnostics.py: a private coercion helper, which the wic020 tests cover.
  • test_settings_boundary.py: the converter's required arguments.
• Benchmark harness. test_benchmarks.py (4) and its CI step tested the weekly timing script, not Sophios. benchmarks_weekly.yml still runs the script and publishes the table, and a broken harness now turns that run red.

Known gaps

• The release workflow is no longer checked on each PR, only when release-please cuts a release.

[darkclad]

LGTM