feat(generators): add --normalize-prefixes flag for well-known prefix names - #3446

2026-10-23, second push (review): (1) the hand-written override table is gone: where prefixmaps' names differ from the names rdflib binds by default (geosparql/geo, sc/schema, WGS84/wgs), rdflib's win, read from Graph(bind_namespaces="rdflib"); the resulting map is identical (16,440 entries). (2) The option was added to every generator's command line and silently ignored by most; it is now declared only on gen-owl, gen-shacl, gen-jsonld-context and gen-jsonld, which honour it, with tests that the others reject it. (3) The docs no longer call the mapping a static table. Still blocked on a prefixmaps release (below). test_generators + test_scripts + test_prefixes + test_sqlite.py: 2093 passed, 54 skipped; the one local failure, test_datadict_personinfo (metamodel version line), fails on main locally too.

2026-10-23: rebased onto main (a1593d16c) as a single commit again (the GitHub merge commit ff4883929 is gone). Conflicts in packages/linkml/pyproject.toml (the pydantic and pyjsg bumps next to the prefixmaps git pin) and uv.lock, which was regenerated from main's lock with uv lock; only the prefixmaps entries differ from main. No code change. The Docker job fails on this branch because the image has no git for the prefixmaps pin; it passes once prefixmaps publishes the fix and the pin becomes a version bound.

"prefixmaps @ git+https://github.com/linkml/prefixmaps@7543515",

ERROR: Cannot find command 'git' - do you have 'git' installed and in your PATH?
ERROR: process "/bin/sh -c pip install /tmp/linkml-whl/*.whl" did not complete
       successfully: exit code: 1

Summary

Adds an opt-in --normalize-prefixes flag to gen-owl, gen-shacl, gen-jsonld-context and gen-jsonld, which forwards it to its inline context. The flag renames non-standard prefix aliases for well-known namespaces to their conventional names, for example sdo: → schema: and dce: → dc:. Namespace IRIs never change; only prefix labels do.

Where the names come from

well_known_prefix_map() derives the names from prefixmaps: its curated linked_data context takes precedence over its broader merged context. Where rdflib binds a namespace by default, its name wins, because rdflib serializes the output: geo, schema and wgs where prefixmaps says geosparql, sc and WGS84. There is no hand-written table. Both http://schema.org/ (from prefixmaps) and https://schema.org/ (from rdflib) map to schema. The pin explained above is needed for the gx correction in prefixmaps#82.

Behaviour

• OWL and SHACL share normalize_graph_prefixes(). It renames schema-declared aliases first, then runtime-injected bindings, and never overwrites an HTTPS binding with its HTTP variant.
• A conventional name that is already declared for a different namespace is left alone, with a warning.
• The JSON-LD context renames both its prefix declarations and the CURIEs it emits.
• The flag defaults to off, so existing output is unchanged.
• Only the generators that honour it offer it. gen-rdf writes the schema's class_uri and prefix declarations as literal data, and gen-shex's RDF already uses rdflib's names, so neither emits an alias to rename.

How was this tested?

test_normalize_prefixes.py (27 tests) and three tests in test_jsonldcontextgen.py cover each generator, the HTTP/HTTPS variants, collisions, and agreement between the generators. The full compiler suite passes after rebasing onto main (12,421 passed).

test_csv_limit now restores csv.field_size_limit. It left the limit at 20, so prefixmaps' CSV loading failed in any later test on the same pytest-xdist worker. That produced intermittent failures, reproducible by running test_csv_limit[20-1] before these tests.

Checklist

• My code follows the contributor guidelines
• I have added tests that prove my fix/feature works
• Existing tests pass locally with my changes