Summarize streams with real progress: adapters reassemble SSE/NDJSON deltas and report 0-99% as tokens arrive

backend/shonar/services/ai/_llm.py

@@ -38,6 +38,30 @@ def build_user_message(transcript: str, title: str | None,
     return msg
 
 
+# Rough typical length of a summary JSON reply. The stream cannot know the
+# model's final size, so progress is an estimate that crawls toward 99 and
+# the job sets 100 on success — honest "almost there", never a fake jump.
+EST_OUTPUT_CHARS = 1200
+
+
+def make_progress_ticker(on_progress):
+    """Wrap an optional on_progress callback into a feed(n_chars) sink.
+
+    Reports a clamped 0..99 estimate from streamed output size. Silent
+    no-op when the caller has no callback; failures never disturb the
+    summary itself."""
+    if on_progress is None:
+        return lambda n_chars: None
+
+    def feed(n_chars: int) -> None:
+        try:  # noqa: SIM105 — swallow deliberately: progress is display state
+            on_progress(min(99, n_chars * 100 // EST_OUTPUT_CHARS))
+        except Exception:  # display state only
+            pass
+
+    return feed
+
+
 def parse_summary(data: object, model: str) -> SummaryResult:
     if not isinstance(data, dict):
         return SummaryResult(model=model)