Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

fix(news): skip source rows whose url is not a string (#5900) (#5934)

LearningCircuita4d99170334b
diff --git a/changelog.d/+news-feed-non-string-source-url.bugfix.md b/changelog.d/+news-feed-non-string-source-url.bugfix.md
new file mode 100644
index 0000000000..06a9fbec50
--- /dev/null
+++ b/changelog.d/+news-feed-non-string-source-url.bugfix.md
@@ -0,0 +1 @@
+The news feed no longer fails with a generic "Database error" when a stored source row's URL is not a string; such rows are now skipped instead of aborting the whole feed.
diff --git a/src/local_deep_research/news/api.py b/src/local_deep_research/news/api.py
index 2a7fef002c..3217e8a349 100644
--- a/src/local_deep_research/news/api.py
+++ b/src/local_deep_research/news/api.py
@@ -429,10 +429,20 @@ def get_news_feed(
)
except Exception as db_error:
- logger.exception(f"Database error in research history: {db_error}")
+ # This block is wide enough to catch far more than database
+ # errors -- everything the feed does with a row runs inside
+ # it. Name the concrete exception type and chain it, so a
+ # plain ValueError from a formatting helper is not reported
+ # to the logs as "Database error" with no way back to the
+ # real raise site. The client still gets the generic detail.
+ logger.exception(
+ "Error in research history query ({}): {}",
+ type(db_error).__name__,
+ db_error,
+ )
raise DatabaseAccessException(
"research_history_query", _GENERIC_ERROR_DETAIL
- )
+ ) from db_error
# If no news items found, return empty list
if not news_items:
diff --git a/src/local_deep_research/web/services/report_assembly_service.py b/src/local_deep_research/web/services/report_assembly_service.py
index 343e893c0d..5b16f042c6 100644
--- a/src/local_deep_research/web/services/report_assembly_service.py
+++ b/src/local_deep_research/web/services/report_assembly_service.py
@@ -170,14 +170,25 @@ def _format_source_link(row: ResearchResource) -> Optional[Dict[str, str]]:
Shared by :func:`get_research_source_links` and its batched variant so
the two cannot drift. Returns ``None`` for a row the feed cannot link
- to (a non-``http`` URL). Titles fall back to the bare domain when
- missing and are truncated to 50 chars, matching the existing
- list-card rendering in the news UI.
+ to: a non-``http`` URL, or a ``url`` that is not a string at all.
+ Titles fall back to the bare domain when missing and are truncated to
+ 50 chars, matching the existing list-card rendering in the news UI.
+
+ A non-``str`` ``url`` is skipped, not coerced — the same rule
+ ``count_distinct_sources`` and ``format_links_to_markdown`` already
+ apply, and for the same reason: the caller passes this result to
+ :func:`_dedup_key`, and ``canonical_url_key`` raises on a non-str
+ (it unpacks ``str.partition``). Coercing with ``str()`` instead would
+ render a Python repr as a clickable link. ``title`` gets the same
+ treatment, so a malformed row degrades to the domain fallback rather
+ than printing a repr onto a news card.
"""
- url = (row.url or "").strip()
+ if not isinstance(row.url, str):
+ return None
+ url = row.url.strip()
if not url.startswith("http"):
return None
- title = (row.title or "").strip()
+ title = (row.title if isinstance(row.title, str) else "").strip()
if not title:
domain = url.split("//")[-1].split("/")[0]
title = domain.replace("www.", "")
diff --git a/tests/web/services/test_report_assembly_service.py b/tests/web/services/test_report_assembly_service.py
index 9250362465..523da024dc 100644
--- a/tests/web/services/test_report_assembly_service.py
+++ b/tests/web/services/test_report_assembly_service.py
@@ -1,6 +1,8 @@
"""Tests for the report_assembly_service module."""
from datetime import datetime, UTC
+from types import SimpleNamespace
+from unittest.mock import MagicMock
import pytest
from sqlalchemy import create_engine
@@ -526,3 +528,97 @@ class TestGetResearchSourceLinksBatch:
)
batch = get_research_source_links_batch([r.id], db_session, limit=2)
assert len(batch[r.id]) == 2
+
+
+# ---------------------------------------------------------------------------
+# Non-string row values (regression: #5900 / #5894)
+# ---------------------------------------------------------------------------
+
+
+class _FakeQuery:
+ """Minimal stand-in for the query chain both source-link helpers build.
+
+ A real SQLite session cannot carry these rows: the point of the tests
+ below is a ``url`` that is not a string, which the column would coerce
+ or reject before the code under test ever saw it.
+ """
+
+ def __init__(self, rows):
+ self._rows = rows
+
+ def filter(self, *args, **kwargs):
+ return self
+
+ def filter_by(self, *args, **kwargs):
+ return self
+
+ def order_by(self, *args, **kwargs):
+ return self
+
+ def limit(self, *args, **kwargs):
+ return self
+
+ def all(self):
+ return list(self._rows)
+
+ def __iter__(self):
+ # get_research_source_links walks the query lazily rather than
+ # calling .all(), so both access shapes must work.
+ return iter(self._rows)
+
+
+class _FakeSession:
+ def __init__(self, rows):
+ self._rows = rows
+
+ def query(self, *args, **kwargs):
+ return _FakeQuery(self._rows)
+
+
+def _row(url, title="T", research_id="r1"):
+ return SimpleNamespace(research_id=research_id, url=url, title=title)
+
+
+class TestSourceLinksNonStringValues:
+ """A row whose ``url`` is not a ``str`` must be skipped, never raise.
+
+ ``_dedup_key`` -> ``canonical_url_key`` -> ``_parse_library_citation``
+ unpacks ``str.partition("#")`` into three names. Handed a non-str that
+ merely *looks* truthy and string-like, that unpack raises
+ ``ValueError: not enough values to unpack (expected 3, got 0)``. In the
+ news feed this happened inside a blanket ``except Exception`` that
+ re-raised it as a ``DatabaseAccessException``, hiding the real defect.
+ """
+
+ def test_magicmock_row_is_skipped_not_fatal(self):
+ """The exact shape the news API tests supply.
+
+ ``mock_session.query(...)`` is stubbed once and reused, so
+ ``get_research_source_links_batch`` receives auto-specced
+ ``MagicMock`` rows whose ``.url`` responds truthily to ``strip``
+ and ``startswith`` but is not a string.
+ """
+ row = MagicMock()
+ row.research_id = "r1"
+ # .url and .title deliberately left as auto-created attributes.
+ batch = get_research_source_links_batch(["r1"], _FakeSession([row]))
+ assert batch == {"r1": []}
+
+ @pytest.mark.parametrize("bad_url", [12345, None, object(), b"https://x"])
+ def test_non_string_url_skipped_good_rows_survive(self, bad_url):
+ rows = [_row(bad_url), _row("https://ok.com/x", title="OK")]
+ batch = get_research_source_links_batch(["r1"], _FakeSession(rows))
+ assert batch["r1"] == [{"url": "https://ok.com/x", "title": "OK"}]
+
+ def test_non_string_title_falls_back_to_domain(self):
+ """A bad title must not put a repr on a news card."""
+ batch = get_research_source_links_batch(
+ ["r1"], _FakeSession([_row("https://ok.com/x", title=object())])
+ )
+ assert batch["r1"] == [{"url": "https://ok.com/x", "title": "ok.com"}]
+
+ def test_single_research_variant_guards_too(self):
+ """The unbatched helper shares ``_format_source_link``; assert it."""
+ rows = [_row(MagicMock()), _row("https://ok.com/x", title="OK")]
+ links = get_research_source_links("r1", _FakeSession(rows), limit=3)
+ assert links == [{"url": "https://ok.com/x", "title": "OK"}]