BUG: fix json_normalize with meta far different from record_path

pandas/io/json/_normalize.py

@@ -533,19 +533,33 @@ def _pull_records(js: dict[str, Any], spec: list | str) -> list:
     meta_vals: DefaultDict = defaultdict(list)
     meta_keys = [sep.join(val) for val in _meta]
 
-    def _recursive_extract(data, path, seen_meta, level: int = 0) -> None:
+    def _recursive_meta_extract(data, meta_remaining_path, meta_key, seen_meta) -> None:
+        if isinstance(data, dict):
+            if len(meta_remaining_path) > 1:
+                if meta_remaining_path[0] in data:
+                    _recursive_meta_extract(data[meta_remaining_path[0]], meta_remaining_path[1:], meta_key, seen_meta)
+                else:
+                    if errors == "ignore":
+                        seen_meta[meta_key] = np.nan
+                    raise KeyError(f"SubKey {meta_remaining_path[0]} of key {meta_key} not found in meta data.")
+            else:
+                seen_meta[meta_key] = _pull_field(data, meta_remaining_path[0])
+
+    def _recursive_extract(data, passed_path, remaining_path, seen_meta, level: int = 0) -> None:
         if isinstance(data, dict):
             data = [data]
-        if len(path) > 1:
+        if len(remaining_path) > 1:
             for obj in data:
                 for val, key in zip(_meta, meta_keys):
-                    if level + 1 == len(val):
+                    if level + 1 == len(val) and passed_path == val[:-1]:
                         seen_meta[key] = _pull_field(obj, val[-1])
+                    elif level + 1 < len(val) and passed_path == val[:level] and remaining_path[0] != val[level]:
+                        _recursive_meta_extract(obj, val[level:], key, seen_meta)
 
-                _recursive_extract(obj[path[0]], path[1:], seen_meta, level=level + 1)
+                _recursive_extract(obj[remaining_path[0]], passed_path + remaining_path[:1], remaining_path[1:], seen_meta, level=level + 1)
         else:
             for obj in data:
-                recs = _pull_records(obj, path[0])
+                recs = _pull_records(obj, remaining_path[0])
                 recs = [
                     nested_to_record(r, sep=sep, max_level=max_level)
                     if isinstance(r, dict)
@@ -556,14 +570,14 @@ def _recursive_extract(data, path, seen_meta, level: int = 0) -> None:
                 # For repeating the metadata later
                 lengths.append(len(recs))
                 for val, key in zip(_meta, meta_keys):
-                    if level + 1 > len(val):
+                    if key in seen_meta:
                         meta_val = seen_meta[key]
                     else:
                         meta_val = _pull_field(obj, val[level:])
                     meta_vals[key].append(meta_val)
                 records.extend(recs)
 
-    _recursive_extract(data, record_path, {}, level=0)
+    _recursive_extract(data, [], record_path, {}, level=0)
 
     result = DataFrame(records)
 

pandas/tests/io/json/test_normalize.py

@@ -301,6 +301,44 @@ def test_shallow_nested(self):
         expected = DataFrame(ex_data, columns=result.columns)
         tm.assert_frame_equal(result, expected)
 
+    def c(self):
+        data = [
+            {
+                "state": "Florida",
+                "shortname": "FL",
+                "info": {"governor": "Rick Scott"},
+                "details": {"counties": [
+                        {"name": "Dade", "population": 12345},
+                        {"name": "Broward", "population": 40000},
+                        {"name": "Palm Beach", "population": 60000},
+                    ]
+                },
+            },
+            {
+                "state": "Ohio",
+                "shortname": "OH",
+                "info": {"governor": "John Kasich"},
+                "details": {"counties": [
+                        {"name": "Summit", "population": 1234},
+                        {"name": "Cuyahoga", "population": 1337},
+                    ]
+                },
+            },
+        ]
+
+        result = json_normalize(
+            data, ["details", "counties"], ["state", "shortname", ["info", "governor"]]
+        )
+        ex_data = {
+            "name": ["Dade", "Broward", "Palm Beach", "Summit", "Cuyahoga"],
+            "state": ["Florida"] * 3 + ["Ohio"] * 2,
+            "shortname": ["FL", "FL", "FL", "OH", "OH"],
+            "info.governor": ["Rick Scott"] * 3 + ["John Kasich"] * 2,
+            "population": [12345, 40000, 60000, 1234, 1337],
+        }
+        expected = DataFrame(ex_data, columns=result.columns)
+        tm.assert_frame_equal(result, expected)
+
     def test_nested_meta_path_with_nested_record_path(self, state_data):
         # GH 27220
         result = json_normalize(