Serialize subject and predicate literals correctly

[joachimvh]

Closes #329

This does have as disadvantage that invalid turtle will be generated if quads get added to the writer that can not be represented in turtle. If that is a problem, an alternative solution would be to have variants of the writer, similarly to how n-quads get handled, where the turtle version throws an error if it encounters a literal in the subject/predicate position and the N3 version handles it correctly. Let me know what you prefer and I can update the PR.

[RubenVerborgh]

invalid turtle will be generated if quads get added to the writer that can not be represented in turtle

I wonder if this has repercussions for #339; should we have a generic flag for "out of syntax features"?

[RubenVerborgh]

Perhaps we should still keep _encodeSubject and _encodeObject around, just so we can error in those functions as needed.

[jeswr]

I wonder if this has repercussions for #339; should we have a generic flag for "out of syntax features"?

...perhaps, though I worry about performance degradation doing it internally; we could just supply validTurtleQuad, validN3Quad etc. functions that consumers can optionally pass their data through first to validate.

[jeswr]

As an aside - are there any RDF syntaxes that don't support a shortand in the subject or object position? If not then the following logic could just be moved to _encodeTerm

N3.js/src/N3Writer.js

Line 207 in c7540c4

return predicate.value === rdf.type ? 'a' : this._encodeTerm(predicate);

[joachimvh]

As an aside - are there any RDF syntaxes that don't support a shortand in the subject or object position?

Turtle only allows it in the predicate position.

[jeswr]

Per @jeswr, I've pushed directly to this branch (thanks to "allow edits by maintainers") to bring it up to date — @joachimvh's diagnosis and fix are untouched and remain the core of this PR; his original commit is preserved as-is.

What was added on top:

• Merged current main into the branch (merge, not rebase). Conflicts were limited to _encodeQuad (now RDF 1.2 <<(...)>> triple-term syntax) and the writer tests.
• Per @RubenVerborgh's earlier suggestion, _encodeSubject and _encodeObject are kept as thin wrappers around the shared literal-aware _encodeTerm, so they remain entry points for format-dependent errors if we want them later.
• quadToString predicates now also go through the literal-aware encoding.
• Additional tests: datatyped and language-tagged literal subjects, literal predicates, text/n3 output, quadToString variants, and an N3 round-trip.

npm run lint and npm test pass with 100% coverage. Should be ready for @jeswr's re-review.

[TallTed]

@jeswr commented ... "Per @jeswr, I've pushed ... Should be ready for @jeswr's re-review."

How many @jeswr are there?