Skip to content

Release/1.7.7 - #49

Merged
ThHanke merged 13 commits into
mainfrom
release/1.7.7
Oct 16, 2026
Merged

ThHanke merged 13 commits into
mainfrom
release/1.7.7

Conversation

@ThHanke

@ThHanke ThHanke commented Oct 16, 2026

Copy link
Copy Markdown
Owner

Changelog

[1.7.7] — 2026-10-16

Reasoning

  • rdf-reasoner-konclude 0.8.0. The reasoner digests the store content on every call, rebuilds only when it changed, recycles its own Worker when the WASM heap grows, and recovers from stalled kernels. It also fixes lossy realization, a cardinality crash and an owl:AsymmetricProperty self-loop gap.
  • One reasoner across runs. The per-run terminate/recreate workaround is gone; an unchanged graph costs no rebuild. The entailment-explanation cache is keyed on the reasoning input, so edits invalidate it.
  • Stop button terminates in-flight reasoning (play button becomes a stop button in the top bar and reasoning indicator).
  • Benchmarks rerun on 0.8.0 (headless Chromium, median of 3): reasoning-demo 1.1 s, Pizza 1.4 s, PMDCO 2.6 s, GALEN 3.4 s, LUBM-1 65.6 s. On one page with PMDCO loaded, a run after a one-individual edit takes about 0.3 s once the first run has started the reasoner (2.3 s), and an unchanged graph about 0.5 s.

Bug fixes

  • Unload / graph removal now drop urn:vg:inferred: inferences no longer outlive the axioms they came from.
  • Canvas: unloading an ontology no longer re-adds its subjects as canvas elements (the P boxes with "No annotation properties").
  • SPARQL: bare UPDATE patterns are scoped to urn:vg:data, and a delete always refreshes the canvas.
  • Relay: no isProcessing deadlock or lost batch calls on timeout; hard maximum call deadline with duplicate-call visibility; no duplicate-call toast for chat history entries; notifiedDupSigs seeded correctly from page history; JSON extraction restarts after a malformed { instead of aborting.

New features

WebMCP support (document.modelContext)

  • Full WebMCP spec compliance: document.modelContext polyfill, object-form registerTool, toolchange events, and signal-based unregistration
  • TypeScript types for the polyfill and WebMCP surface
  • Test suite covering polyfill registration, toolchange dispatch, and signal cleanup

AI Relay Bridge — Gemini support

  • Input detection for Gemini's rich-textarea contenteditable component
  • Send-button detection via aria-label="Send message" with visibility guard (Gemini hides the send button with a CSS hidden class during generation instead of using disabled)
  • Streaming detection covers both "Stop generating" and "Stop response" button label variants

AI Relay Bridge — reliability improvements

  • Session locking: relay popup binds to a specific Ontosphere tab; stale-tab messages are dropped
  • Fast session discovery via ping rather than waiting for the next poll cycle
  • 3-minute batch timeout with partial-result flush when individual tool calls hang
  • Incremental injection with pending-count hints for multi-call batches
  • Injection guard blocks new polls while a result is being inserted, preventing dropped results
  • FhGenie stop-button streaming detection (DismissSquare icon path)
  • getGraphState scoped by graph list; unified with canvas summary

SHACL violations

  • Violation messages enriched with rdfs:label values for human-readable property and class names
  • Improved tooltip readability in the SHACL shapes panel

Bug fixes

  • Canvas: Delete key and halo delete button now route through the authoring state machine; nodes no longer reappear after save when deleted via keyboard
  • Relay: reverted polling strategy to isAiStreaming() guard after stability regression; added two-poll idle streak requirement before injection to prevent the UUID-echo bug on OWUI
  • Relay: strict bidirectional session pairing blocks tabs running old bookmarklet code

Breaking changes

None. The WebMCP polyfill is opt-in and backwards-compatible with the existing window.__mcpTools registration path.

Thomas Hanke added 13 commits October 16, 2026 14:12
When reasoning runs, the play button (▶/▶) is replaced by a stop button
(■) in both TopBar and ReasoningIndicator. Clicking stop sends a
stopReasoning command to the worker, which calls resetDlReasoner() to
terminate the Konclude/WASM inner reasoner. The pending runReasoning
promise rejects, the finally block resets isReasoning=false, and the
play button reappears. RDF graph data in the outer worker is unaffected.

New worker command: stopReasoning (protocol + runtime + workerClient +
impl). New handler: handleStopReasoning in ReactodiaCanvas.
…eout

Two bugs fixed:

1. resetCallTimeout() was called on every vg-ping, including the relay
   popup's own 10 s heartbeat (no sessionId). When Ontosphere stops
   responding, these heartbeats endlessly reset the 30 s deadline →
   isProcessing stays true forever → idlePoll never scans for new calls.
   Fix: only extend the deadline on Ontosphere-sourced pings (sessionId
   present), i.e. when the app is provably still alive.

2. When a per-call or batch timeout fires, callQueue = [] discarded
   remaining calls whose signatures were already in dispatchedSigs, making
   them invisible to idlePoll forever. Fix: un-mark those sigs before
   clearing the queue so the next idlePoll cycle can re-extract them from
   the page. Same cleanup applied when the popup fails to open.

Also adds pendingSig tracking so the currently-dispatched call's sig is
also removed from dispatchedSigs on timeout, allowing the AI to retry an
identical call.

Also stages relay-mock-chat.html (Claude.ai TipTap mock mode) which was
modified but not committed in the 1.7.6 release.
Two fixes for the scenario where an AI retries a failing tool call:

1. Hard max deadline (CALL_HARD_MAX_MS = 90 s): even when Ontosphere
   keeps pinging with sessionId (tool is hung on network), resetCallTimeout
   now caps the next timer at (hardMax - elapsed) so the call always times
   out within 90 s from dispatch, not infinitely.

2. Duplicate-call toast: when idlePoll finds tool calls in the page but
   all are already in dispatchedSigs (AI retried identical id+params), a
   visible warning toast now fires at most once per 10 s explaining that
   the relay ignored them and asking the AI to change the call. Previously
   this was silent, leaving the user with no feedback.

extractAllToolCalls now returns null sentinels for already-seen calls
(counted separately), and idlePoll uses filterNew/countSkipped helpers
so the dispatch path only sees genuinely new calls.
The previous toast used a 10 s time debounce which still fired repeatedly
because idlePoll scans the full page on every cycle and old chat-history
calls are permanently visible. Replaced with a notifiedDupSigs Set: a
toast fires at most once per unique duplicate sig for the lifetime of the
bookmarklet instance. Pre-seed also absorbs existing dups silently so no
spurious toast fires on load.
The previous pre-seed called newDupSigs() on _preSeed, but _preSeed
contains fresh call objects (isDup is unset) because dispatchedSigs was
empty at that point. So notifiedDupSigs stayed empty, and the very first
idlePoll cycle would find all history calls as fresh dups and fire the
toast for them all.

Fix: directly add each pre-seed call's sig to notifiedDupSigs so history
calls are silently absorbed. Toast only fires for sigs that appear for
the first time AFTER the bookmarklet loads.
… breaking

A JSON-RPC call with an unquoted key (e.g. {Jsiirpc"...}) phase-shifts
the string-boundary tracker in extractJsonObjects: the inner { chars get
absorbed as phantom-string content, depth never reaches 0, and the
function returns complete=false → break, silently discarding every valid
tool call that follows on the same page.

Change the break to i = start + 1; continue so the scanner skips the
bad opening brace and picks up the next { — the AI's corrected retry call
is then extracted and dispatched normally.

Regression test added: malformed call followed by valid retry now yields
exactly one dispatched call.
…t after delete

Two bugs caused SPARQL DELETE/INSERT via queryGraph to appear to succeed
but leave the canvas unchanged:

1. Graph targeting: Comunica follows the SPARQL 1.1 spec — bare patterns
   in DELETE/INSERT/WHERE without an explicit GRAPH clause target the
   default graph (empty in our N3 store), not urn:vg:data. Fix: after
   parsing, BGP patterns in update operations that have no GRAPH wrapper
   are automatically scoped to GRAPH <urn:vg:data>. Explicit GRAPH clauses
   are left untouched.

2. Empty canvas refresh: after a complete DELETE the subject set is empty,
   so the if (emission.subjects.length > 0) guard suppressed emitSubjects,
   leaving all nodes on the canvas. Fix: always call emitSubjects after a
   SPARQL update so the canvas clears correctly.
… reasoner across runs

0.8.0 digests store content on every call and rebuilds only on change, so the
per-run terminate/recreate workaround and the reset on graph removal are gone.
Entailment cache is now keyed on the reasoning input generation.
… graph removed

Inferred quads outlive the axioms they came from; their orphaned subjects
reappeared on the canvas after unloading an ontology.
The unload emission lists the removed subjects, which the incremental sync
treated as new and created elements for. Drop them from the emission and
from knownSubjects, and never create elements on unload.
Also restores the Table 2 benchmark runner, updated for the current MCP tool
exposure, with a repeated-run mode.
@ThHanke
ThHanke merged commit f2712cc into main Oct 16, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant