jsonld.expand incorrect when there are identical scoped contexts

[rossbowen]

I have an example of a context where two of my terms have an identical scoped context - here creator and publisher have a scoped context which happens to be identical.

import json
from rdflib import Graph
from pyld import jsonld

CONTEXT = {
    "dcat": "http://www.w3.org/ns/dcat#",
    "dcterms": "http://purl.org/dc/terms/",
    "creator": {
        "@id": "dcterms:creator",
        "@type": "@id",
        "@context": {
            "@base": "http://example.org/organisations/",
        },
    },
    "publisher": {
        "@id": "dcterms:publisher",
        "@type": "@id",
        "@context": {
            "@base": "http://example.org/organisations/",
        },
    },
}

FRAME = {
    "@context": CONTEXT,
    "@type": "dcat:DatasetSeries",
}

db = Graph()
db.parse(
    data="""
        @prefix dcat: <http://www.w3.org/ns/dcat#> .
        @prefix dcterms: <http://purl.org/dc/terms/> .

        <http://example.org/datasets/cpih> a dcat:DatasetSeries ;
            dcterms:publisher <http://example.org/organisations/acme-inc> ;
            dcterms:creator <http://example.org/organisations/acme-inc> ;
            .
         """
)

result = db.query(
    """
        PREFIX dcat: <http://www.w3.org/ns/dcat#>
        PREFIX dcterms: <http://purl.org/dc/terms/>
        CONSTRUCT {
            ?ds a dcat:DatasetSeries ;
                dcterms:publisher ?publisher ;
                dcterms:creator ?creator ;
                .
        }
        WHERE {
            ?ds a dcat:DatasetSeries ;
                dcterms:publisher ?publisher ;
                dcterms:creator ?creator ;
                .
        }
    """
)

ld = json.loads(result.serialize(format="json-ld", context=CONTEXT))
framed = jsonld.frame(ld, FRAME)

My expected output is:

{
  "@context": {
    "dcat": "http://www.w3.org/ns/dcat#",
    "dcterms": "http://purl.org/dc/terms/",
    "creator": {
      "@id": "dcterms:creator",
      "@type": "@id",
      "@context": { "@base": "http://example.org/organisations/" }
    },
    "publisher": {
      "@id": "dcterms:publisher",
      "@type": "@id",
      "@context": { "@base": "http://example.org/organisations/" }
    }
  },
  "@id": "http://example.org/datasets/cpih",
  "@type": "dcat:DatasetSeries",
  "creator": "acme-inc",
  "publisher": "acme-inc"
}

But instead I get (notice the dcterms:publisher and the full URI):

{
  "@context": {
    "dcat": "http://www.w3.org/ns/dcat#",
    "dcterms": "http://purl.org/dc/terms/",
    "creator": {
      "@id": "dcterms:creator",
      "@type": "@id",
      "@context": { "@base": "http://example.org/organisations/" }
    },
    "publisher": {
      "@id": "dcterms:publisher",
      "@type": "@id",
      "@context": { "@base": "http://example.org/organisations/" }
    }
  },
  "@id": "http://example.org/datasets/cpih",
  "@type": "dcat:DatasetSeries",
  "creator": "acme-inc",
  "dcterms:publisher": "http://example.org/organisations/acme-inc"
}

I've stepped through the code a bit and noticed the issue isn't with the framing, but with the expansion.

Running JsonLdProcessor().expand(input_=ld, options=None) provides the following - notice how publisher gets an @value instead of a @id.

[
  {
    "@id": "http://example.org/datasets/cpih",
    "@type": ["http://www.w3.org/ns/dcat#DatasetSeries"],
    "http://purl.org/dc/terms/creator": [
      { "@id": "http://example.org/organisations/acme-inc" }
    ],
    "http://purl.org/dc/terms/publisher": [
      { "@value": "http://example.org/organisations/acme-inc" }
    ]
  }
]

Stepping through it further, the issue seems to be with the _process_context() not differentiating in the case of having two identical scoped contexts, which results in published being dropped from the term_ctx mapping.

pyld/lib/pyld/jsonld.py

Lines 2679 to 2684 in 316fbc2

	# use potential scoped context for key
	term_ctx = active_ctx
	ctx = JsonLdProcessor.get_context_value(active_ctx, key, '@context')
	if ctx is not None:
	term_ctx = self._process_context(active_ctx, ctx, options,
	propagate=True, override_protected=True)

A workaround for now is to trivially make the contexts different from one another.

[rossbowen]

An issue looks to be here:

pyld/lib/pyld/jsonld.py

Lines 5510 to 5515 in 316fbc2

	if base and '@base' in active_ctx:
	# The None case preserves rval as potentially relative
	if active_ctx['@base'] is not None:
	rval = prepend_base(prepend_base(base, active_ctx['@base']), rval)
	elif base:
	rval = prepend_base(base, rval)

base is usually passed though as an empty string "" so these conditions do not trigger.

Fixing returns this:

input = {
    "@context": {
        "a": {
            "@id": "http://example.org/a",
            "@type": "@id",
            "@context": {
                "@base": "http://example.org/",
            },
        },
    },
    "@id": "http://example.org/ex",
    "a": "a1",
}

expected = [
    {
        "@id": "http://example.org/ex",
        "http://example.org/a": [{"@id": "http://example.org/a1"}],
    }
]

jsonld.expand(input) == expected