feat(all): enable passing in-memory data to create_table

.github/workflows/ibis-backends.yml

@@ -141,6 +141,7 @@ jobs:
             extras:
               - mysql
               - geospatial
+              - polars
             sys-deps:
               - libgeos-dev
           - name: postgres
@@ -186,6 +187,7 @@ jobs:
             title: MS SQL Server
             extras:
               - mssql
+              - polars
             services:
               - mssql
             sys-deps:
@@ -216,6 +218,7 @@ jobs:
             serial: true
             extras:
               - oracle
+              - polars
             services:
               - oracle
           - name: flink
@@ -271,6 +274,7 @@ jobs:
               extras:
                 - mysql
                 - geospatial
+                - polars
               services:
                 - mysql
               sys-deps:
@@ -352,6 +356,7 @@ jobs:
               title: MS SQL Server
               extras:
                 - mssql
+                - polars
               services:
                 - mssql
               sys-deps:
@@ -381,6 +386,7 @@ jobs:
               serial: true
               extras:
                 - oracle
+                - polars
               services:
                 - oracle
           - os: ubuntu-latest

ibis/backends/bigquery/__init__.py

@@ -13,7 +13,6 @@
 import google.auth.credentials
 import google.cloud.bigquery as bq
 import google.cloud.bigquery_storage_v1 as bqstorage
-import pandas as pd
 import pydata_google_auth
 import sqlglot as sg
 import sqlglot.expressions as sge
@@ -42,6 +41,8 @@
     from collections.abc import Callable, Iterable, Mapping
     from pathlib import Path
 
+    import pandas as pd
+    import polars as pl
     import pyarrow as pa
     from google.cloud.bigquery.table import RowIterator
 
@@ -940,7 +941,12 @@
     def create_table(
         self,
         name: str,
-        obj: pd.DataFrame | pa.Table | ir.Table | None = None,
+        obj: ir.Table
+        | pd.DataFrame
+        | pa.Table
+        | pl.DataFrame
+        | pl.LazyFrame
+        | None = None,
         *,
         schema: ibis.Schema | None = None,
         database: str | None = None,
@@ -1027,14 +1033,11 @@
             for name, value in (options or {}).items()
         )
 
-        if obj is not None:
-            import pyarrow as pa
-            import pyarrow_hotfix  # noqa: F401
+        if obj is not None and not isinstance(obj, ir.Table):
+            obj = ibis.memtable(obj, schema=schema)
 
-            if isinstance(obj, (pd.DataFrame, pa.Table)):
-                obj = ibis.memtable(obj, schema=schema)
-
-            self._register_in_memory_tables(obj)
+        # This is a no-op if there aren't any memtables
+        self._register_in_memory_tables(obj)
 
         if temp:
             dataset = self._session_dataset.dataset_id

ibis/backends/clickhouse/__init__.py

@@ -34,6 +34,7 @@
     from pathlib import Path
 
     import pandas as pd
+    import polars as pl
 
 
 def _to_memtable(v):
@@ -586,7 +587,12 @@ def read_csv(
     def create_table(
         self,
         name: str,
-        obj: pd.DataFrame | pa.Table | ir.Table | None = None,
+        obj: ir.Table
+        | pd.DataFrame
+        | pa.Table
+        | pl.DataFrame
+        | pl.LazyFrame
+        | None = None,
         *,
         schema: ibis.Schema | None = None,
         database: str | None = None,

ibis/backends/datafusion/__init__.py

@@ -14,7 +14,6 @@
 import sqlglot as sg
 import sqlglot.expressions as sge
 
-import ibis
 import ibis.common.exceptions as com
 import ibis.expr.datatypes as dt
 import ibis.expr.operations as ops
@@ -24,6 +23,7 @@
 from ibis.backends.datafusion.compiler import DataFusionCompiler
 from ibis.backends.sql import SQLBackend
 from ibis.backends.sql.compiler import C
+from ibis.common.dispatch import lazy_singledispatch
 from ibis.expr.operations.udf import InputType
 from ibis.formats.pyarrow import PyArrowType
 from ibis.util import gen_name, normalize_filename
@@ -40,6 +40,7 @@
 
 if TYPE_CHECKING:
     import pandas as pd
+    import polars as pl
 
 
 class Backend(SQLBackend, CanCreateCatalog, CanCreateDatabase, CanCreateSchema, NoUrl):
@@ -272,7 +273,13 @@
         list[str]
             The list of the table names that match the pattern `like`.
         """
-        return self._filter_with_like(self.con.tables(), like)
+        database = database or "public"
+        query = (
+            sg.select("table_name")
+            .from_("information_schema.tables")
+            .where(sg.column("table_schema").eq(sge.convert(database)))
+        )
+        return self.raw_sql(query).to_pydict()["table_name"]
 
     def get_schema(
         self,
@@ -550,7 +557,14 @@
     def create_table(
         self,
         name: str,
-        obj: pd.DataFrame | pa.Table | ir.Table | None = None,
+        obj: ir.Table
+        | pd.DataFrame
+        | pa.Table
+        | pa.RecordBatchReader
+        | pa.RecordBatch
+        | pl.DataFrame
+        | pl.LazyFrame
+        | None = None,
         *,
         schema: sch.Schema | None = None,
         database: str | None = None,
@@ -589,12 +603,10 @@
 
         quoted = self.compiler.quoted
 
-        if obj is not None:
-            if not isinstance(obj, ir.Expr):
-                table = ibis.memtable(obj)
-            else:
-                table = obj
+        if isinstance(obj, ir.Expr):
+            table = obj
 
+            # If it's a memtable, it will get registered in the pre-execute hooks
             self._run_pre_execute_hooks(table)
 
             relname = "_"
@@ -610,10 +622,13 @@
                     sg.to_identifier(relname, quoted=quoted)
                 )
             )
+        elif obj is not None:
+            _read_in_memory(obj, name, self, overwrite=overwrite)
+            return self.table(name, database=database)
         else:
             query = None
 
-        table_ident = sg.to_identifier(name, quoted=quoted)
+        table_ident = sg.table(name, db=database, quoted=quoted)
 
         if query is None:
             column_defs = [
@@ -670,3 +685,92 @@
         ident = sg.table(name, db=db, catalog=catalog).sql(self.name)
         with self._safe_raw_sql(sge.delete(ident)):
             pass
+
+
+@contextlib.contextmanager
+def _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+    """Workaround inability to overwrite tables in dataframe API.
+
+    Datafusion has helper methods for loading in-memory data, but these methods
+    don't allow overwriting tables.
+    The SQL interface allows creating tables from existing tables, so we register
+    the data as a table using the dataframe API, then run a
+
+    CREATE [OR REPLACE] TABLE table_name AS SELECT * FROM in_memory_thing
+
+    and that allows us to toggle the overwrite flag.
+    """
+    src = sge.Create(
+        this=table_name,
+        kind="TABLE",
+        expression=sg.select("*").from_(tmp_name),
+        replace=overwrite,
+    )
+
+    yield
+
+    _conn.raw_sql(src)
+    _conn.drop_table(tmp_name)
+
+
+@lazy_singledispatch
+def _read_in_memory(
+    source: Any, table_name: str, _conn: Backend, overwrite: bool = False
+):
+    raise NotImplementedError("No support for source or imports missing")
+
+
+@_read_in_memory.register(dict)
+def _pydict(source, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("pydict")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.from_pydict(source, name=tmp_name)
+
+
+@_read_in_memory.register("polars.DataFrame")
+def _polars(source, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("polars")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.from_polars(source, name=tmp_name)
+
+
+@_read_in_memory.register("polars.LazyFrame")
+def _polars(source, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("polars")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.from_polars(source.collect(), name=tmp_name)
+
+
+@_read_in_memory.register("pyarrow.Table")
+def _pyarrow_table(source, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("pyarrow")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.from_arrow_table(source, name=tmp_name)
+
+
+@_read_in_memory.register("pyarrow.RecordBatchReader")
+def _pyarrow_rbr(source, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("pyarrow")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.from_arrow_table(source.read_all(), name=tmp_name)
+
+
+@_read_in_memory.register("pyarrow.RecordBatch")
+def _pyarrow_rb(source, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("pyarrow")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.register_record_batches(tmp_name, [[source]])
+
+
+@_read_in_memory.register("pyarrow.dataset.Dataset")
+def _pyarrow_rb(source, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("pyarrow")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.register_dataset(tmp_name, source)
+
+
+@_read_in_memory.register("pandas.DataFrame")
+def _pandas(source: pd.DataFrame, table_name, _conn, overwrite: bool = False):
+    tmp_name = gen_name("pandas")
+    with _create_and_drop_memtable(_conn, table_name, tmp_name, overwrite):
+        _conn.con.from_pandas(source, name=tmp_name)

ibis/backends/duckdb/__init__.py

@@ -28,12 +28,14 @@
 from ibis.backends.duckdb.converter import DuckDBPandasData
 from ibis.backends.sql import SQLBackend
 from ibis.backends.sql.compiler import STAR, C
+from ibis.common.dispatch import lazy_singledispatch
 from ibis.expr.operations.udf import InputType
 
 if TYPE_CHECKING:
     from collections.abc import Iterable, Mapping, MutableMapping, Sequence
 
     import pandas as pd
+    import polars as pl
     import torch
     from fsspec import AbstractFileSystem
 
@@ -121,7 +123,12 @@ def _to_sqlglot(
     def create_table(
         self,
         name: str,
-        obj: pd.DataFrame | pa.Table | ir.Table | None = None,
+        obj: ir.Table
+        | pd.DataFrame
+        | pa.Table
+        | pl.DataFrame
+        | pl.LazyFrame
+        | None = None,
         *,
         schema: ibis.Schema | None = None,
         database: str | None = None,
@@ -846,11 +853,19 @@ def _read_parquet_pyarrow_dataset(
         # explicitly.
 
     def read_in_memory(
+        # TODO: deprecate this in favor of `create_table`
         self,
-        source: pd.DataFrame | pa.Table | pa.ipc.RecordBatchReader,
+        source: pd.DataFrame
+        | pa.Table
+        | pa.RecordBatchReader
+        | pl.DataFrame
+        | pl.LazyFrame,
         table_name: str | None = None,
     ) -> ir.Table:
-        """Register a Pandas DataFrame or pyarrow object as a table in the current database.
+        """Register an in-memory table object in the current database.
+
+        Supported objects include pandas DataFrame, a Polars
+        DataFrame/LazyFrame, or a PyArrow Table or RecordBatchReader.
 
         Parameters
         ----------
@@ -867,13 +882,7 @@ def read_in_memory(
 
         """
         table_name = table_name or util.gen_name("read_in_memory")
-        self.con.register(table_name, source)
-
-        if isinstance(source, pa.ipc.RecordBatchReader):
-            # Ensure the reader isn't marked as started, in case the name is
-            # being overwritten.
-            self._record_batch_readers_consumed[table_name] = False
-
+        _read_in_memory(source, table_name, self)
         return self.table(table_name)
 
     def read_delta(
@@ -1598,3 +1607,28 @@ def _get_temp_view_definition(self, name: str, definition: str) -> str:
     def _create_temp_view(self, table_name, source):
         with self._safe_raw_sql(self._get_temp_view_definition(table_name, source)):
             pass
+
+
+@lazy_singledispatch
+def _read_in_memory(source: Any, table_name: str, _conn: Backend, **kwargs: Any):
+    raise NotImplementedError(
+        f"The `{_conn.name}` backend currently does not support "
+        f"reading data of {type(source)!r}"
+    )
+
+
+@_read_in_memory.register("polars.DataFrame")
+@_read_in_memory.register("polars.LazyFrame")
+@_read_in_memory.register("pyarrow.Table")
+@_read_in_memory.register("pandas.DataFrame")
+@_read_in_memory.register("pyarrow.dataset.Dataset")
+def _default(source, table_name, _conn, **kwargs: Any):
+    _conn.con.register(table_name, source)
+
+
+@_read_in_memory.register("pyarrow.RecordBatchReader")
+def _pyarrow_rbr(source, table_name, _conn, **kwargs: Any):
+    _conn.con.register(table_name, source)
+    # Ensure the reader isn't marked as started, in case the name is
+    # being overwritten.
+    _conn._record_batch_readers_consumed[table_name] = False