test(python): cover null keys in merge insert

docs/src/js/README.md

@@ -34,36 +34,6 @@ const results = await table.vectorSearch([0.1, 0.3]).limit(20).toArray();
 console.log(results);
 ```
 
-### Use an Existing Table with LangChain
-
-When wrapping an existing table with `@langchain/community`, open the table
-with LanceDB and pass the resulting table handle to LangChain. The LangChain
-`uri` and `tableName` options are used when creating a table; they do not open
-an existing table for search.
-
-```javascript
-import { LanceDB as LangChainLanceDB } from "@langchain/community/vectorstores/lancedb";
-import * as lancedb from "@lancedb/lancedb";
-
-const db = await lancedb.connect("data/sample-lancedb");
-const table = await db.openTable("my_table");
-const vectorStore = new LangChainLanceDB(embeddings, {
-  table,
-  textKey: "item",
-});
-
-const documents = await vectorStore.similaritySearch("foo", 5);
-```
-
-Affected versions of `@langchain/community` do not map LanceDB's `_distance`
-column to the score returned by `similaritySearchVectorWithScore`. Query the
-table directly when you need the numeric vector distance:
-
-```javascript
-const results = await table.vectorSearch(queryVector).limit(5).toArray();
-console.log(results[0]._distance);
-```
-
 The [quickstart](https://docs.lancedb.com/quickstart/) contains more complete examples.
 
 ## Development

nodejs/README.md

@@ -30,36 +30,6 @@ const results = await table.vectorSearch([0.1, 0.3]).limit(20).toArray();
 console.log(results);
 ```
 
-### Use an Existing Table with LangChain
-
-When wrapping an existing table with `@langchain/community`, open the table
-with LanceDB and pass the resulting table handle to LangChain. The LangChain
-`uri` and `tableName` options are used when creating a table; they do not open
-an existing table for search.
-
-```javascript
-import { LanceDB as LangChainLanceDB } from "@langchain/community/vectorstores/lancedb";
-import * as lancedb from "@lancedb/lancedb";
-
-const db = await lancedb.connect("data/sample-lancedb");
-const table = await db.openTable("my_table");
-const vectorStore = new LangChainLanceDB(embeddings, {
-  table,
-  textKey: "item",
-});
-
-const documents = await vectorStore.similaritySearch("foo", 5);
-```
-
-Affected versions of `@langchain/community` do not map LanceDB's `_distance`
-column to the score returned by `similaritySearchVectorWithScore`. Query the
-table directly when you need the numeric vector distance:
-
-```javascript
-const results = await table.vectorSearch(queryVector).limit(5).toArray();
-console.log(results[0]._distance);
-```
-
 The [quickstart](https://docs.lancedb.com/quickstart/) contains more complete examples.
 
 ## Development

python/python/tests/test_table.py

@@ -2341,6 +2341,45 @@ def test_merge_insert(mem_db: DBConnection):
         )
 
 
+def test_merge_insert_with_null_on_column(mem_db: DBConnection):
+    table = mem_db.create_table(
+        "users",
+        data=pa.table(
+            {
+                "id": [0, 1],
+                "name": ["Alice", "Bob"],
+                "record_type": [None, "personal"],
+            }
+        ),
+    )
+    new_data = pa.table(
+        {
+            "id": [1, 2],
+            "name": ["Bobby", "Charlie"],
+            "record_type": ["personal", None],
+        }
+    )
+
+    result = (
+        table.merge_insert(["id", "record_type"])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(new_data)
+    )
+
+    assert result.num_inserted_rows == 1
+    assert result.num_updated_rows == 1
+    assert result.num_deleted_rows == 0
+    expected = pa.table(
+        {
+            "id": [0, 1, 2],
+            "name": ["Alice", "Bobby", "Charlie"],
+            "record_type": [None, "personal", None],
+        }
+    )
+    assert table.to_arrow().sort_by("id") == expected
+
+
 def test_merge_insert_by_source_delete_expr(mem_db: DBConnection):
     table = mem_db.create_table(
         "my_table",