Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

fix(parser): resolve PDF bookmark destinations for integer page indices (#794)

Matt Van Horna1c4a05bdb3a
diff --git a/openviking/parse/parsers/pdf.py b/openviking/parse/parsers/pdf.py
index 9fb197437..382213e99 100644
--- a/openviking/parse/parsers/pdf.py
+++ b/openviking/parse/parsers/pdf.py
@@ -261,8 +261,9 @@ class PDFParser(BaseParser):
# Group bookmarks by page_num
bookmarks_by_page = defaultdict(list)
for bm in bookmarks:
- if bm["page_num"]:
- bookmarks_by_page[bm["page_num"]].append(bm)
+ # Fall back to page 1 for unresolvable destinations
+ page = bm["page_num"] or 1
+ bookmarks_by_page[page].append(bm)
for page_num, page in enumerate(pdf.pages, 1):
# Inject headings before page text
@@ -314,10 +315,7 @@ class PDFParser(BaseParser):
f"Failed to extract image {img_idx + 1} on page {page_num}: {img_err}"
)
- # Append any bookmarks with unresolved page numbers at the end
- unresolved = [bm for bm in bookmarks if bm.get("page_num") is None]
- if unresolved:
- logger.debug(f"{len(unresolved)} bookmarks had unresolved page numbers")
+ # Note: bookmarks with unresolvable page numbers are injected at page 1
if not parts:
logger.warning(f"No content extracted from {pdf_path}")
@@ -372,6 +370,11 @@ class PDFParser(BaseParser):
resolved = page_ref.resolve()
if hasattr(resolved, "objid"):
page_num = objid_to_num.get(resolved.objid)
+ elif isinstance(page_ref, int):
+ # 0-based integer page index (common in many PDF producers)
+ candidate = page_ref + 1
+ if 1 <= candidate <= len(pdf.pages):
+ page_num = candidate
except Exception:
pass
@@ -415,7 +418,7 @@ class PDFParser(BaseParser):
[
s
for s, count in size_counter.items()
- if s >= body_size + min_delta and count < size_counter[body_size] * 0.3
+ if s >= body_size + min_delta and count < size_counter[body_size] * 0.5
],
reverse=True,
)
diff --git a/tests/parse/test_pdf_bookmark_extraction.py b/tests/parse/test_pdf_bookmark_extraction.py
index aa672d8a0..ea0fbcc1b 100644
--- a/tests/parse/test_pdf_bookmark_extraction.py
+++ b/tests/parse/test_pdf_bookmark_extraction.py
@@ -105,6 +105,49 @@ class TestExtractBookmarks:
assert len(bookmarks) == 1
assert bookmarks[0]["page_num"] is None
+ def test_extract_bookmarks_integer_page_index(self):
+ """Bookmarks with integer destination (0-based) are resolved correctly."""
+ mock_pdf = MagicMock()
+
+ mock_page1 = MagicMock()
+ mock_page1.page_obj.objid = 100
+ mock_page2 = MagicMock()
+ mock_page2.page_obj.objid = 200
+ mock_pdf.pages = [mock_page1, mock_page2]
+
+ # Integer page indices instead of object references
+ mock_pdf.doc.get_outlines.return_value = [
+ (1, "Chapter 1", [0, "/Fit"], None, None),
+ (1, "Chapter 2", [1, "/Fit"], None, None),
+ ]
+
+ bookmarks = self.parser._extract_bookmarks(mock_pdf)
+ assert len(bookmarks) == 2
+ assert bookmarks[0]["page_num"] == 1
+ assert bookmarks[0]["title"] == "Chapter 1"
+ assert bookmarks[1]["page_num"] == 2
+ assert bookmarks[1]["title"] == "Chapter 2"
+
+ def test_extract_bookmarks_integer_page_index_out_of_range(self):
+ """Out-of-range integer page indices are treated as unresolved."""
+ mock_pdf = MagicMock()
+
+ mock_page1 = MagicMock()
+ mock_page1.page_obj.objid = 100
+ mock_pdf.pages = [mock_page1] # Only 1 page
+
+ mock_pdf.doc.get_outlines.return_value = [
+ (1, "Valid", [0, "/Fit"], None, None),
+ (1, "Too High", [5, "/Fit"], None, None),
+ (1, "Negative", [-1, "/Fit"], None, None),
+ ]
+
+ bookmarks = self.parser._extract_bookmarks(mock_pdf)
+ assert len(bookmarks) == 3
+ assert bookmarks[0]["page_num"] == 1
+ assert bookmarks[1]["page_num"] is None
+ assert bookmarks[2]["page_num"] is None
+
def test_extract_bookmarks_exception_returns_empty(self):
"""Returns empty list on unexpected exceptions (best-effort)."""
mock_pdf = MagicMock()