ENH: Implementing rowid

docs/source/api.rst

@@ -397,6 +397,7 @@ Table methods
    TableExpr.mutate
    TableExpr.projection
    TableExpr.relabel
+   TableExpr.rowid
    TableExpr.schema
    TableExpr.set_column
    TableExpr.sort_by

docs/source/release/index.rst

@@ -12,6 +12,7 @@ Release Notes
    These release notes are for versions of ibis **1.0 and later**. Release
    notes for pre-1.0 versions of ibis can be found at :doc:`release-pre-1.0`
 
+* :feature:`2251` Add ``rowid`` expression, supported by SQLite and OmniSciDB
 * :feature:`2230` Add intersection to general ibis api
 * :support:`2304` Update ``google-cloud-bigquery`` dependency minimum version to 1.12.0
 * :feature:`2303` Add ``application_name`` argument to ``ibis.bigquery.connect`` to allow attributing Google API requests to projects that use Ibis.

ibis/expr/api.py

@@ -4278,6 +4278,32 @@ def _table_drop(self, fields):
     return self[[field for field in schema if field not in field_set]]
 
 
+def _rowid(self):
+    """
+    An autonumeric representing the row number of the results.
+
+    It can be 0 or 1 indexed depending on the backend. Check the backend
+    documentation.
+
+    Note that this is different from the window function row number
+    (even if they are conceptually the same), and different from row
+    id in backends where it represents the physical location (e.g. Oracle
+    or PostgreSQL's ctid).
+
+    Returns
+    -------
+    ir.IntegerColumn
+
+    Examples
+    --------
+    >>> my_table[my_table.rowid(), my_table.name].execute()
+    1|Ibis
+    2|pandas
+    3|Dask
+    """
+    return ops.RowID().to_expr()
+
+
 _table_methods = dict(
     aggregate=aggregate,
     count=_table_count,
@@ -4308,6 +4334,7 @@ def _table_drop(self, fields):
     union=_table_union,
     intersect=_table_intersect,
     view=_table_view,
+    rowid=_rowid,
 )
 
 

ibis/expr/operations.py

@@ -249,6 +249,19 @@ def _make_expr(self):
         return klass(self, name=self.name)
 
 
+class RowID(ValueOp):
+    """The row number (an autonumeric) of the returned result."""
+
+    def output_type(self):
+        return dt.int64.column_type()
+
+    def resolve_name(self):
+        return 'rowid'
+
+    def has_resolved_name(self):
+        return True
+
+
 def find_all_base_tables(expr, memo=None):
     if memo is None:
         memo = {}

ibis/omniscidb/operations.py

@@ -1085,6 +1085,7 @@ def formatter(translator, expr):
     ops.IsNan: unary('isNan'),
     ops.NullIfZero: _nullifzero,
     ops.ZeroIfNull: _zeroifnull,
+    ops.RowID: lambda *args: 'rowid',
 }
 
 # WINDOW

ibis/pandas/execution/generic.py

@@ -1066,3 +1066,10 @@ def execute_simple_case_series(op, value, whens, thens, otherwise, **kwargs):
 @execute_node.register(ops.Distinct, pd.DataFrame)
 def execute_distinct_dataframe(op, df, **kwargs):
     return df.drop_duplicates()
+
+
+@execute_node.register(ops.RowID)
+def execute_rowid(op, *args, **kwargs):
+    raise com.UnsupportedOperationError(
+        'rowid is not supported in pandas backends'
+    )

ibis/sql/sqlite/compiler.py

@@ -285,6 +285,7 @@ def _rpad(t, expr):
         ops.StandardDev: toolz.compose(
             sa.func._ibis_sqlite_sqrt, _variance_reduction('_ibis_sqlite_var')
         ),
+        ops.RowID: lambda t, expr: sa.literal_column('rowid'),
     }
 )
 

ibis/tests/all/test_column.py

@@ -1,6 +1,11 @@
+import numpy as np
+import pandas as pd
 import pytest
 
 
+ROWID_ZERO_INDEXED_BACKENDS = ('omniscidb',)
+
+
 @pytest.mark.parametrize(
     'column',
     [
@@ -16,3 +21,29 @@ def test_distinct_column(backend, alltypes, df, column):
     result = expr.execute()
     expected = df[column].unique()
     assert set(result) == set(expected)
+
+
+@pytest.mark.xfail_unsupported
+def test_rowid(con, backend):
+    t = con.table('functional_alltypes')
+    result = t[t.rowid()].execute()
+    first_value = 0 if backend.name in ROWID_ZERO_INDEXED_BACKENDS else 1
+    expected = pd.Series(
+        range(first_value, first_value + len(result)),
+        dtype=np.int64,
+        name='rowid',
+    )
+    pd.testing.assert_series_equal(result.iloc[:, 0], expected)
+
+
+@pytest.mark.xfail_unsupported
+def test_named_rowid(con, backend):
+    t = con.table('functional_alltypes')
+    result = t[t.rowid().name('number')].execute()
+    first_value = 0 if backend.name in ROWID_ZERO_INDEXED_BACKENDS else 1
+    expected = pd.Series(
+        range(first_value, first_value + len(result)),
+        dtype=np.int64,
+        name='number',
+    )
+    pd.testing.assert_series_equal(result.iloc[:, 0], expected)