fix: don't use v2 by default on empty table (#1469)

python/python/tests/test_db.py

@@ -552,6 +552,12 @@ async def test_create_in_v2_mode(tmp_path):
 
     assert await is_in_v2_mode(tbl)
 
+    # Create empty table uses v1 mode by default
+    tbl = await db.create_table("test_empty_v2_default", data=None, schema=schema)
+    await tbl.add(make_table())
+
+    assert not await is_in_v2_mode(tbl)
+
 
 def test_replace_index(tmp_path):
     db = lancedb.connect(uri=tmp_path)

rust/lancedb/src/connection.rs

@@ -220,7 +220,7 @@ impl CreateTableBuilder<false, NoData> {
             mode: CreateTableMode::default(),
             write_options: WriteOptions::default(),
             embeddings: Vec::new(),
-            use_legacy_format: false,
+            use_legacy_format: true,
         }
     }
 

rust/lancedb/src/table.rs

@@ -1079,7 +1079,10 @@ impl NativeTable {
         params: Option<WriteParams>,
         read_consistency_interval: Option<std::time::Duration>,
     ) -> Result<Self> {
-        let params = params.unwrap_or_default();
+        let params = params.unwrap_or(WriteParams {
+            use_legacy_format: true,
+            ..Default::default()
+        });
         // patch the params if we have a write store wrapper
         let params = match write_store_wrapper.clone() {
             Some(wrapper) => params.patch_with_store_wrapper(wrapper)?,
@@ -1625,11 +1628,16 @@ impl TableInternal for NativeTable {
         }
 
         // patch the params if we have a write store wrapper
-        let lance_params = match self.store_wrapper.clone() {
+        let mut lance_params = match self.store_wrapper.clone() {
             Some(wrapper) => lance_params.patch_with_store_wrapper(wrapper)?,
             None => lance_params,
         };
 
+        // Only use the new format if the user passes use_legacy_format=False in while creating
+        // a table with data.  We don't want to accidentally switch to v2 format during an add
+        // operation.  If the table is already v2 this won't have any effect.
+        lance_params.use_legacy_format = true;
+
         self.dataset.ensure_mutable().await?;
         let dataset = Dataset::write(data, &self.uri, Some(lance_params)).await?;