SPARQL example: authentication, named graphs, and recent connections in the connection dialog

[kvistgaard]

SPARQL example: authentication, named graphs, and recent connections in the connection dialog

Summary

Extends the connection dialog in the sparql example (the playground) so an
endpoint that is not open and unrestricted can be explored without writing code:

• HTTP Basic authentication — optional username/password, sent as an
  Authorization header via the existing
  SparqlDataProviderOptions.queryFunction extension point. No library changes.
• Named graph restriction — one or more graph IRIs, applied as SPARQL 1.1
  Protocol default-graph-uri parameters (§2.1.4), so all queries see the merge
  of the listed graphs instead of the whole dataset.
• Recent connections — each successful connection (endpoint, graphs,
  username) is remembered in localStorage, listed in the dialog, and refills
  the form on click. Up to 8, deduplicated, removable.

Motivation for the graph restriction

Exploring a whole endpoint stops working once the endpoint is large: on
knowledge graphs with billions of triples the class tree and statistics queries
aggregate over everything the store holds, which is slow at best and a timeout
at worst — and the result mixes every dataset on the endpoint. Restricting the
dataset to named graphs makes such endpoints usable from the playground:

• exploration scopes to one dataset on a shared endpoint that hosts many;
• class counts and link statistics describe the dataset of interest instead of
  the union of unrelated graphs;
• a data graph can be combined with its ontology graph when the schema (class
  and property declarations, labels, hierarchy) is stored separately — without
  it there are no link types to display;
• no server-side configuration is needed, where the alternative is a dedicated
  restricted endpoint per dataset.

Handling of credentials

• Passwords are kept in tab-scoped sessionStorage only: they survive a reload,
  die with the tab, and never enter the URL hash or the recent-connections list.
• Stored credentials are keyed to the endpoint they were entered for and are
  attached only when the hash names that endpoint, so a pasted link to a
  different endpoint can never receive them.
• Clicking a recent connection that used a username focuses the (deliberately
  empty) password field, so the missing value is visible before Connect.
• The endpoint and graph IRIs stay in the URL hash as before, so a configuration
  remains a bookmarkable link. The hash is now written once per save and parsed
  from location.href rather than location.hash, whose getter returns a
  percent-decoded value in Firefox and would corrupt endpoint URLs containing
  encoded & or =. Graph IRIs split on whitespace only, since a comma is a
  legal IRI character.

Testing

Driven end-to-end in headless Chromium against a mock endpoint requiring Basic
auth (every request carried the header and the graph parameters, before and
after reload, and credentials stored for one endpoint are not attached to
another) and against a Virtuoso 08.03.3332 endpoint holding an 11M-triple named
graph (restriction verified by comparing type-instance counts with and without
the graph parameters). npm run lint, npm run typecheck, and the examples
build pass.

🤖 Generated with Claude Code