# -*- coding: utf-8 -*- # https://github.com/RDFLib/rdflib-jsonld/blob/feature/json-ld-1.1/rdflib_jsonld/util.py import typing as t if t.TYPE_CHECKING: import json else: try: import json assert json # workaround for pyflakes issue #13 except ImportError: import simplejson as json from os import sep from os.path import normpath from urllib.parse import urljoin, urlsplit, urlunsplit from rdflib.parser import create_input_source from io import StringIO def source_to_json(source): # TODO: conneg for JSON (fix support in rdflib's URLInputSource!) source = create_input_source(source, format="json-ld") stream = source.getByteStream() try: return json.load(StringIO(stream.read().decode("utf-8"))) finally: stream.close() VOCAB_DELIMS = ("#", "/", ":") def split_iri(iri): for delim in VOCAB_DELIMS: at = iri.rfind(delim) if at > -1: return iri[: at + 1], iri[at + 1 :] return iri, None def norm_url(base, url): """ >>> norm_url('http://example.org/', '/one') 'http://example.org/one' >>> norm_url('http://example.org/', '/one#') 'http://example.org/one#' >>> norm_url('http://example.org/one', 'two') 'http://example.org/two' >>> norm_url('http://example.org/one/', 'two') 'http://example.org/one/two' >>> norm_url('http://example.org/', 'http://example.net/one') 'http://example.net/one' >>> norm_url('http://example.org/', 'http://example.org//one') 'http://example.org//one' """ if "://" in url: return url parts = urlsplit(urljoin(base, url)) path = normpath(parts[2]) if sep != "/": path = "/".join(path.split(sep)) if parts[2].endswith("/") and not path.endswith("/"): path += "/" result = urlunsplit(parts[0:2] + (path,) + parts[3:]) if url.endswith("#") and not result.endswith("#"): result += "#" return result def context_from_urlinputsource(source): if source.content_type == "application/json": # response_info was added to InputSource in rdflib 4.2 try: links = source.response_info.getallmatchingheaders("Link") except AttributeError: return for link in links: if ' rel="http://www.w3.org/ns/json-ld#context"' in link: i, j = link.index("<"), link.index(">") if i > -1 and j > -1: return urljoin(source.url, link[i + 1 : j])