feat: schema evolution APIs in all SDKs (#1851)

* Support `add_columns`, `alter_columns`, `drop_columns` in Remote SDK
and async Python
* Add `data_type` parameter to node
* Docs updates

nodejs/__test__/table.test.ts

@@ -825,6 +825,18 @@ describe("schema evolution", function () {
       new Field("price", new Float64(), true),
     ]);
     expect(await table.schema()).toEqual(expectedSchema);
+
+    await table.alterColumns([{ path: "new_id", dataType: "int32" }]);
+    const expectedSchema2 = new Schema([
+      new Field("new_id", new Int32(), true),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float32(), true)),
+        true,
+      ),
+      new Field("price", new Float64(), true),
+    ]);
+    expect(await table.schema()).toEqual(expectedSchema2);
   });
 
   it("can drop a column from the schema", async function () {

nodejs/examples/basic.test.ts

@@ -116,6 +116,26 @@ test("basic table examples", async () => {
       await tbl.add(data);
       // --8<-- [end:add_data]
     }
+
+    {
+      // --8<-- [start:add_columns]
+      await tbl.addColumns([{ name: "double_price", valueSql: "price * 2" }]);
+      // --8<-- [end:add_columns]
+      // --8<-- [start:alter_columns]
+      await tbl.alterColumns([
+        {
+          path: "double_price",
+          rename: "dbl_price",
+          dataType: "float",
+          nullable: true,
+        },
+      ]);
+      // --8<-- [end:alter_columns]
+      // --8<-- [start:drop_columns]
+      await tbl.dropColumns(["dbl_price"]);
+      // --8<-- [end:drop_columns]
+    }
+
     {
       // --8<-- [start:vector_search]
       const res = await tbl.search([100, 100]).limit(2).toArray();

nodejs/src/table.rs

@@ -178,16 +178,20 @@ impl Table {
     #[napi(catch_unwind)]
     pub async fn alter_columns(&self, alterations: Vec<ColumnAlteration>) -> napi::Result<()> {
         for alteration in &alterations {
-            if alteration.rename.is_none() && alteration.nullable.is_none() {
+            if alteration.rename.is_none()
+                && alteration.nullable.is_none()
+                && alteration.data_type.is_none()
+            {
                 return Err(napi::Error::from_reason(
-                    "Alteration must have a 'rename' or 'nullable' field.",
+                    "Alteration must have a 'rename', 'dataType', or 'nullable' field.",
                 ));
             }
         }
         let alterations = alterations
             .into_iter()
-            .map(LanceColumnAlteration::from)
-            .collect::<Vec<_>>();
+            .map(LanceColumnAlteration::try_from)
+            .collect::<std::result::Result<Vec<_>, String>>()
+            .map_err(napi::Error::from_reason)?;
 
         self.inner_ref()?
             .alter_columns(&alterations)
@@ -433,24 +437,43 @@ pub struct ColumnAlteration {
     /// The new name of the column. If not provided then the name will not be changed.
     /// This must be distinct from the names of all other columns in the table.
     pub rename: Option<String>,
+    /// A new data type for the column. If not provided then the data type will not be changed.
+    /// Changing data types is limited to casting to the same general type. For example, these
+    /// changes are valid:
+    /// * `int32` -> `int64` (integers)
+    /// * `double` -> `float` (floats)
+    /// * `string` -> `large_string` (strings)
+    /// But these changes are not:
+    /// * `int32` -> `double` (mix integers and floats)
+    /// * `string` -> `int32` (mix strings and integers)
+    pub data_type: Option<String>,
     /// Set the new nullability. Note that a nullable column cannot be made non-nullable.
     pub nullable: Option<bool>,
 }
 
-impl From<ColumnAlteration> for LanceColumnAlteration {
-    fn from(js: ColumnAlteration) -> Self {
+impl TryFrom<ColumnAlteration> for LanceColumnAlteration {
+    type Error = String;
+    fn try_from(js: ColumnAlteration) -> std::result::Result<Self, Self::Error> {
         let ColumnAlteration {
             path,
             rename,
             nullable,
+            data_type,
         } = js;
-        Self {
+        let data_type = if let Some(data_type) = data_type {
+            Some(
+                lancedb::utils::string_to_datatype(&data_type)
+                    .ok_or_else(|| format!("Invalid data type: {}", data_type))?,
+            )
+        } else {
+            None
+        };
+        Ok(Self {
             path,
             rename,
             nullable,
-            // TODO: wire up this field
-            data_type: None,
-        }
+            data_type,
+        })
     }
 }