codex/ingest.py

@@ -220,13 +220,18 @@ def ingest_paper(
 
         if formulas or figures:
             with get_conn() as conn2:
+                with conn2.cursor() as cur:
+                    # Delete-before-insert keeps re-ingest idempotent (BIGSERIAL has no
+                    # natural UNIQUE key, so ON CONFLICT DO NOTHING never fires).
+                    cur.execute("DELETE FROM formulas WHERE paper_id = %s", (paper.id,))
+                    cur.execute("DELETE FROM figures  WHERE paper_id = %s", (paper.id,))
+
                 if formulas:
                     with conn2.cursor() as cur:
                         cur.executemany(
                             """
                             INSERT INTO formulas (paper_id, page, raw_latex, context, eq_label)
                             VALUES (%s, %s, %s, %s, %s)
-                            ON CONFLICT DO NOTHING
                             """,
                             [
                                 (f.paper_id, f.page, f.raw_latex, f.context, f.eq_label)
@@ -241,7 +246,6 @@ def ingest_paper(
                             """
                             INSERT INTO figures (paper_id, page, caption, image_path)
                             VALUES (%s, %s, %s, %s)
-                            ON CONFLICT DO NOTHING
                             """,
                             [
                                 (fig.paper_id, fig.page, fig.caption, fig.image_path)

codex/parsing/figures.py

@@ -126,6 +126,7 @@ def extract_figures(
         logger.exception("fitz.open failed for %s", pdf_path)
         return results
 
+    try:
         for page_num in range(len(doc)):
             page = doc[page_num]
             images: list[Any] = page.get_images(full=True)
@@ -162,5 +163,7 @@ def extract_figures(
                         caption=caption,
                     )
                 )
+    finally:
+        doc.close()
 
     return results

codex/parsing/mathpix.py

@@ -112,6 +112,7 @@ def _extract_formulas_pix2tex(pdf_path: str) -> list[FormulaChunk]:
 
     scale = fitz.Matrix(3, 3)
 
+    try:
         for page_num in range(len(doc)):
             page = doc[page_num]
             blocks: list[Any] = page.get_text("blocks")
@@ -130,7 +131,10 @@ def _extract_formulas_pix2tex(pdf_path: str) -> list[FormulaChunk]:
                     continue
 
                 # Math content filter
-            if _math_symbol_count(text) < _MIN_MATH_SYMBOLS or _math_ratio(text) < _MIN_MATH_RATIO:
+                if (
+                    _math_symbol_count(text) < _MIN_MATH_SYMBOLS
+                    or _math_ratio(text) < _MIN_MATH_RATIO
+                ):
                     continue
 
                 # Crop at 3× for OCR quality
@@ -160,6 +164,8 @@ def _extract_formulas_pix2tex(pdf_path: str) -> list[FormulaChunk]:
                         context=context,
                     )
                 )
+    finally:
+        doc.close()
 
     return results