Save both readings of a menu day when a double read disagrees

parentcalendar/db.py

@@ -7,7 +7,7 @@ from parentcalendar import config
 
 MIGRATIONS = Path(__file__).parent / "migrations"
 JSON_COLUMNS = {"source_ids", "ignore_labels", "image_keys", "items", "flags", "cell_bbox", "before", "after",
-                "child_ids"}  # child_ids comes from json_group_array() subqueries
+                "child_ids", "reads"}  # child_ids comes from json_group_array() subqueries
 BOOL_COLUMNS = {"double_read", "all_day", "applies", "include", "no_lunch"}
 
 

parentcalendar/extraction.py

@@ -464,6 +464,7 @@ def merge_reads(a, b):
             m.setdefault("read_flags", []).append("second read did not find this")
         elif norm("".join(m["items"])) != norm("".join(o["items"])) or m["no_lunch"] != o["no_lunch"]:
             m.setdefault("read_flags", []).append("second read differs: items")
+            m["reads"] = [{"items": r["items"], "no_lunch": r["no_lunch"]} for r in (m, o)]  # review can switch
     for o in menu_b.values():
         o.setdefault("read_flags", []).append("only the second read found this")
         a["menu_items"].append(o)
@@ -534,6 +535,7 @@ def selftest():
                      "No School": ["second read did not find this"],
                      "Happy Halloween": ["only the second read found this"]}, flags
     assert m["menu_items"][0]["read_flags"] == ["second read differs: items"]
+    assert [r["items"] for r in m["menu_items"][0]["reads"]] == [["Corn"], ["Corn", "Peas"]]
     # A word in a stray alphabet is dropped; a page that is really bilingual keeps both
     page = {"events": [], "menu_items": [{"items": ["Chicken Patty", "Corn", "Whole Orange выда"]},
                                          {"items": ["Pizza", "Garden Salad", "Mixed Fruit", "Peas", "Milk"]}]}

parentcalendar/migrations/002_menu_reads.sql

@@ -0,0 +1,3 @@
+-- Both readings of a menu day when a double read disagreed: [{"items": [...], "no_lunch": bool}, {...}].
+-- NULL when the reads agreed or the source reads once. Review lets you switch between them.
+ALTER TABLE menu_item ADD COLUMN reads TEXT;

parentcalendar/scans.py

@@ -113,9 +113,10 @@ def process_scan(conn, scan_id):
             for m in r["menu_items"]:
                 conn.execute(
                     "INSERT OR REPLACE INTO menu_item (scan_id, source_id, date, items, no_lunch, page, cell_bbox,"
-                    " confidence, flags) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)",
+                    " confidence, flags, reads) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
                     (scan_id, source["id"], m["date"], json.dumps(m["items"]), m["no_lunch"], page,
-                     json.dumps(m.get("cell_bbox")), m.get("confidence"), json.dumps(m.get("flags", []))))
+                     json.dumps(m.get("cell_bbox")), m.get("confidence"), json.dumps(m.get("flags", [])),
+                     json.dumps(m["reads"]) if m.get("reads") else None))
         dates = sorted(d for d in dates if d)
         tokens = sum(r["usage"]["input_tokens"] + r["usage"]["output_tokens"] for r in runs)
         conn.execute("UPDATE scan SET status = 'parsed', error = NULL, date_start = ?, date_end = ?, model = ?,"