fix: /documentation NameError from sinatra-contrib 4.2.1 namespace bug

[alexskr]

Summary

• Extracts home-controller helpers out of namespace "/" into a new helpers/home_helper.rb (Sinatra::Helpers::HomeHelper). Fixes the primary bug: /documentation and /metadata/:class raising NameError: undefined local variable or method 'metadata_all' under sinatra-contrib 4.2.1 + mustermann 3.0, because the namespace's before-filter compiles for "/" to a pattern that only matches the root path, so the module that carries the helpers is never mixed into the request instance for sub-paths. Same structural fix agroportal used.
• Drops the dead cardinality column from HomeHelper#metadata_all. The producer (goo's cls.cardinality(attr)) was removed in Refactor: clean model settings module code agroportal/goo#52 and synced into ncbo/goo via Sync: align ncbo/goo with the AgroPortal codebase goo#176 (April 2024); the consumer views never read values[:cardinality] anyway. See issue comment for details.
• Swaps pandoc-ruby for kramdown in the Gemfile. The :markdown filter in views/documentation/documentation.haml was resolving to Tilt's PandocTemplate, which shells out to a pandoc binary that isn't present in our ontoportal/testkit-base image (agroportal installs it explicitly in their Dockerfile; we'd rather avoid a system-binary dep for four prose blocks). kramdown is pure Ruby, already known to Tilt, no compile step, no cross-arch churn.
• Adds a regression test test_documentation_route_renders to test/controllers/test_home_controller.rb so this doesn't silently regress again.

Fixes #212

Bonus: also restores long-broken /metadata/:class endpoints

The same namespace fix revives GET /metadata/:class for every valid LinkedData::Models::* class, e.g.

• https://data.bioontology.org/metadata/Metrics → LinkedData::Models::Metric
• /metadata/Slice, /metadata/ProvisionalClass, /metadata/OntologyType, /metadata/Reply, etc.

These URLs are listed in the long-standing #37 ("broken metadata resources", open since 2017). They've been returning 500s for years — under the prior Sinatra 1.4 stack the specific failure was different, but the user-visible symptom was the same, and the namespace-helper-resolution bug in Sinatra 4.2.1 kept them broken post-migration too. This PR closes out the class-name cases from that issue.

Note: issue #37 also lists property-name paths like /metadata/prefixIRI, /metadata/name, /metadata/omv*. Those aren't valid model class names and previously 500'd with a NameError; this PR makes them return a clean 404 with the message 'X' is not a documented media type. See /documentation or try /metadata/Ontology. (commit d2e5b64). Making /metadata/:name actually render documentation for those attribute URIs — which are legitimately emitted by the JSON-LD serializer in @context — is a separate design question tracked in ncbo/bioportal-project#388.

Related: #37, ncbo/bioportal-project#388

Commits

SHA	Change
b85ee2b	extract 10 helpers + CLASS_MAP into helpers/home_helper.rb; add regression test test_documentation_route_renders
6303343	drop dead cardinality column from HomeHelper#metadata_all (producer removed in goo 2024)
99d0df1	pandoc-ruby → kramdown in Gemfile/Gemfile.lock
5ca5f8d	add regression tests for /metadata/:class (Metrics, Reply)
c37e214	resolve nested model classes in HomeHelper#metadata (sub-module walk for Notes::Reply etc.)
d2e5b64	return 404 from /metadata/:class for non-class names, worded with the BioPortal docs "media type" vocabulary and routed through the error helper
2b105ef	broaden happy-path tests (Ontology, Metrics, Class); narrow 404 test to guaranteed-bogus names so future attribute-documentation work (ncbo/bioportal-project#388) doesn't read as a regression

Test plan

• ruby -c on all changed Ruby files.
• Structural repro (minimal Sinatra 4.2.1 app with the same namespace "/" + helpers Module shape) confirms the NameError is gone — GET /documentation returns 200 where it was 500.
• Post-swap verification: Tilt[:md] resolves to Tilt::KramdownTemplate (not Tilt::PandocTemplate).
• bundle exec ruby -Itest test/controllers/test_home_controller.rb --verbose against the 4store dev backend (via rake 'test:docker:up[fs]') — all 6 tests pass (32 assertions, 0 failures, 0 errors):
  • test_documentation_route_renders
  • test_metadata_route_renders_for_model_class (GET /metadata/Ontology, /metadata/Metrics, /metadata/Class)
  • test_metadata_route_resolves_submodule_class (GET /metadata/Reply → LinkedData::Models::Notes::Reply via the sub-module walk added in c37e214)
  • test_metadata_route_returns_404_for_bogus_names (guaranteed-nonexistent identifiers return 404 with the media type message)
  • test_home_index_returns_links_hash
  • test_home_index_handles_type_uri_failures
• Manual smoke: bundle exec rackup and curl /, /documentation, /metadata/Class, /metadata/Metrics, /metadata/Slice — all 200.

Notes on deviations from agroportal

We adopted agroportal's structural fix (helper module extraction) verbatim but deviated on three secondary points:

Their choice	Ours	Why
Rename metadata_all → get_metadata_all	Keep metadata_all	Cosmetic; would force a view edit for no benefit
Hardcoded routes_cls list in get_metadata_all	Keep dynamic ObjectSpace discovery via routes_by_class	Our GET / and /metadata/:class routes still rely on dynamic route→class mapping; agroportal removed those routes (MOD API catalog serving)
Install pandoc in Dockerfile	Swap gem to kramdown	Avoids a system-binary dep across every container (prod, CI, testkit base); pure-Ruby is a better match for the trivial use case

[alexskr]

Pre-fix review flagged that HomeHelper#metadata(cls) wouldn't resolve nested model classes (e.g. LinkedData::Models::Notes::Reply for /metadata/Reply), which would both break the #37 "bonus" claim and the test_metadata_route_resolves_submodule_class test.

Addressed in 65e4379: extracted a resolve_model_class(name) helper that first tries LinkedData::Models.const_get(name) and, on NameError, falls back to scanning LinkedData::Models.constants for a sub-module that defines the class — mirroring the same sub-module search routes_by_class already does.

Verified against the 4store backend:

TestHomeController#test_metadata_route_resolves_submodule_class   PASS (0.21s)   # GET /metadata/Reply
TestHomeController#test_metadata_route_renders_for_model_class    PASS (0.03s)   # GET /metadata/Metrics
TestHomeController#test_documentation_route_renders               PASS (0.07s)
TestHomeController#test_home_index_returns_links_hash             PASS (0.10s)
TestHomeController#test_home_index_handles_type_uri_failures      PASS (0.00s)
5 tests, 17 assertions, 0 failures, 0 errors

Two small stylistic choices worth noting:

• Extracted the lookup into a named helper rather than inlining in metadata, so the sub-module search logic has one home (future callers or routes_by_class itself could consolidate onto it).
• Iteration is "first wins" rather than routes_by_class's implicit "last wins". No practical difference — no duplicate class names across LinkedData::Models::* sub-modules in the current codebase — but happy to flip it to match routes_by_class exactly if that's preferred.