pixeltable 0.1.2__py3-none-any.whl → 0.2.1__py3-none-any.whl

pixeltable/tests/test_exprs.py

@@ -1,45 +1,50 @@
-import sqlalchemy as sql
+import json
+import urllib.parse
+from typing import List, Dict
+
 import pytest
+import sqlalchemy as sql
 
+import pixeltable as pxt
+import pixeltable.func as func
 from pixeltable import catalog
-from pixeltable.type_system import StringType, BoolType, IntType, ImageType, ArrayType, ColumnType, FloatType
-from pixeltable.function import Function
-from pixeltable.exprs import Expr, CompoundPredicate, FunctionCall, Literal, InlineDict, InlineArray, ColumnRef
+from pixeltable import exceptions as excs
+from pixeltable import exprs
+from pixeltable.exprs import Expr, ColumnRef
 from pixeltable.exprs import RELATIVE_PATH_ROOT as R
-from pixeltable.functions import udf_call, dict_map, cast, sum, count
+from pixeltable.functions import cast, sum, count
 from pixeltable.functions.pil.image import blend
-from pixeltable.functions.clip import encode_image
-from pixeltable import exceptions as exc
-from pixeltable.tests import utils
+from pixeltable.iterators import FrameIterator
+from pixeltable.tests.utils import get_image_files, skip_test_if_not_installed
+from pixeltable.type_system import StringType, BoolType, IntType, ArrayType, ColumnType, FloatType, \
+    VideoType
 
 
 class TestExprs:
-    # This breaks with exception 'cannot pickle _thread._local obj'
-    # sum = Function(
-    #     IntType(), [IntType()],
-    #     init_fn=lambda: TestExprs.SumAggregator(), update_fn=SumAggregator.update, value_fn=SumAggregator.value)
-
     def test_basic(self, test_tbl: catalog.Table) -> None:
         t = test_tbl
-        assert isinstance(t['c1'] < 'a', Expr)
+        assert t['c1'].equals(t.c1)
+        assert t['c7']['*'].f5.equals(t.c7['*'].f5)
+
+        assert isinstance(t.c1 == None, Expr)
         assert isinstance(t.c1 < 'a', Expr)
-        assert isinstance(t['c1'] <= 'a', Expr)
         assert isinstance(t.c1 <= 'a', Expr)
-        assert isinstance(t['c1'] == 'a', Expr)
         assert isinstance(t.c1 == 'a', Expr)
-        assert isinstance(t['c1'] != 'a', Expr)
         assert isinstance(t.c1 != 'a', Expr)
-        assert isinstance(t['c1'] > 'a', Expr)
         assert isinstance(t.c1 > 'a', Expr)
-        assert isinstance(t['c1'] >= 'a', Expr)
         assert isinstance(t.c1 >= 'a', Expr)
         assert isinstance((t.c1 == 'a') & (t.c2 < 5), Expr)
         assert isinstance((t.c1 == 'a') | (t.c2 < 5), Expr)
         assert isinstance(~(t.c1 == 'a'), Expr)
+        with pytest.raises(AttributeError) as excinfo:
+            _ = t.does_not_exist
+        assert 'unknown' in str(excinfo.value).lower()
 
     def test_compound_predicates(self, test_tbl: catalog.Table) -> None:
         t = test_tbl
         # compound predicates that can be fully evaluated in SQL
+        _ = t.where((t.c1 == 'test string') & (t.c6.f1 > 50)).collect()
+        _ = t.where((t.c1 == 'test string') & (t.c2 > 50)).collect()
         e = ((t.c1 == 'test string') & (t.c2 > 50)).sql_expr()
         assert len(e.clauses) == 2
 
@@ -52,46 +57,192 @@ class TestExprs:
         e = (~(t.c1 == 'test string')).sql_expr()
         assert isinstance(e, sql.sql.expression.BinaryExpression)
 
+        with pytest.raises(TypeError) as exc_info:
+            _ = t.where((t.c1 == 'test string') or (t.c6.f1 > 50)).collect()
+        assert 'cannot be used in conjunction with python boolean operators' in str(exc_info.value).lower()
+
         # compound predicates with Python functions
-        udf = Function(BoolType(), [StringType()], eval_fn=lambda a: True)
-        udf2 = Function(BoolType(), [IntType()], eval_fn=lambda a: True)
-
-        # & can be split
-        p = (t.c1 == 'test string') & udf(t.c1)
-        assert p.sql_expr() is None
-        sql_pred, other_pred = p.extract_sql_predicate()
-        assert isinstance(sql_pred, sql.sql.expression.BinaryExpression)
-        assert isinstance(other_pred, FunctionCall)
-
-        p = (t.c1 == 'test string') & udf(t.c1) & (t.c2 > 50)
-        assert p.sql_expr() is None
-        sql_pred, other_pred = p.extract_sql_predicate()
-        assert len(sql_pred.clauses) == 2
-        assert isinstance(other_pred, FunctionCall)
-
-        p = (t.c1 == 'test string') & udf(t.c1) & (t.c2 > 50) & udf2(t.c2)
-        assert p.sql_expr() is None
-        sql_pred, other_pred = p.extract_sql_predicate()
-        assert len(sql_pred.clauses) == 2
-        assert isinstance(other_pred, CompoundPredicate)
-
-        # | cannot be split
-        p = (t.c1 == 'test string') | udf(t.c1)
-        assert p.sql_expr() is None
-        sql_pred, other_pred = p.extract_sql_predicate()
-        assert sql_pred is None
-        assert isinstance(other_pred, CompoundPredicate)
-
-    def test_basic_filter(self, test_tbl: catalog.Table) -> None:
+        @pxt.udf(return_type=BoolType(), param_types=[StringType()])
+        def udf(_: str) -> bool:
+            return True
+        @pxt.udf(return_type=BoolType(), param_types=[IntType()])
+        def udf2(_: int) -> bool:
+            return True
+
+        # TODO: find a way to test this
+        # # & can be split
+        # p = (t.c1 == 'test string') & udf(t.c1)
+        # assert p.sql_expr() is None
+        # sql_pred, other_pred = p.extract_sql_predicate()
+        # assert isinstance(sql_pred, sql.sql.expression.BinaryExpression)
+        # assert isinstance(other_pred, FunctionCall)
+        #
+        # p = (t.c1 == 'test string') & udf(t.c1) & (t.c2 > 50)
+        # assert p.sql_expr() is None
+        # sql_pred, other_pred = p.extract_sql_predicate()
+        # assert len(sql_pred.clauses) == 2
+        # assert isinstance(other_pred, FunctionCall)
+        #
+        # p = (t.c1 == 'test string') & udf(t.c1) & (t.c2 > 50) & udf2(t.c2)
+        # assert p.sql_expr() is None
+        # sql_pred, other_pred = p.extract_sql_predicate()
+        # assert len(sql_pred.clauses) == 2
+        # assert isinstance(other_pred, CompoundPredicate)
+        #
+        # # | cannot be split
+        # p = (t.c1 == 'test string') | udf(t.c1)
+        # assert p.sql_expr() is None
+        # sql_pred, other_pred = p.extract_sql_predicate()
+        # assert sql_pred is None
+        # assert isinstance(other_pred, CompoundPredicate)
+
+    def test_filters(self, test_tbl: catalog.Table) -> None:
         t = test_tbl
         _ = t[t.c1 == 'test string'].show()
         print(_)
         _ = t[t.c2 > 50].show()
         print(_)
+        _ = t[t.c1n == None].show()
+        print(_)
+        _ = t[t.c1n != None].show(0)
+        print(_)
+
+    def test_exception_handling(self, test_tbl: catalog.Table) -> None:
+        t = test_tbl
+
+        # error in expr that's handled in SQL
+        with pytest.raises(excs.Error):
+            _ = t[(t.c2 + 1) / t.c2].show()
+
+        # error in expr that's handled in Python
+        with pytest.raises(excs.Error):
+            _ = t[(t.c6.f2 + 1) / (t.c2 - 10)].show()
+
+        # the same, but with an inline function
+        @pxt.udf(return_type=FloatType(), param_types=[IntType(), IntType()])
+        def f(a: int, b: int) -> float:
+            return a / b
+        with pytest.raises(excs.Error):
+            _ = t[f(t.c2 + 1, t.c2)].show()
+
+        # error in agg.init()
+        @pxt.uda(update_types=[IntType()], value_type=IntType(), name='agg')
+        class Aggregator(pxt.Aggregator):
+            def __init__(self):
+                self.sum = 1 / 0
+            def update(self, val):
+                pass
+            def value(self):
+                return 1
+        with pytest.raises(excs.Error):
+            _ = t[agg(t.c2)].show()
+
+        # error in agg.update()
+        @pxt.uda(update_types=[IntType()], value_type=IntType(), name='agg')
+        class Aggregator(pxt.Aggregator):
+            def __init__(self):
+                self.sum = 0
+            def update(self, val):
+                self.sum += 1 / val
+            def value(self):
+                return 1
+        with pytest.raises(excs.Error):
+            _ = t[agg(t.c2 - 10)].show()
+
+        # error in agg.value()
+        @pxt.uda(update_types=[IntType()], value_type=IntType(), name='agg')
+        class Aggregator(pxt.Aggregator):
+            def __init__(self):
+                self.sum = 0
+            def update(self, val):
+                self.sum += val
+            def value(self):
+                return 1 / self.sum
+        with pytest.raises(excs.Error):
+            _ = t[t.c2 <= 2][agg(t.c2 - 1)].show()
+
+    def test_props(self, test_tbl: catalog.Table, img_tbl: catalog.Table) -> None:
+        t = test_tbl
+        # errortype/-msg for computed column
+        res = t.select(error=t.c8.errortype).collect()
+        assert res.to_pandas()['error'].isna().all()
+        res = t.select(error=t.c8.errormsg).collect()
+        assert res.to_pandas()['error'].isna().all()
+
+        img_t = img_tbl
+        # fileurl
+        res = img_t.select(img_t.img.fileurl).show(0).to_pandas()
+        stored_urls = set(res.iloc[:, 0])
+        assert len(stored_urls) == len(res)
+        all_urls  = set([urllib.parse.urljoin('file:', path) for path in get_image_files()])
+        assert stored_urls <= all_urls
+
+        # localpath
+        res = img_t.select(img_t.img.localpath).show(0).to_pandas()
+        stored_paths = set(res.iloc[:, 0])
+        assert len(stored_paths) == len(res)
+        all_paths  = set(get_image_files())
+        assert stored_paths <= all_paths
+
+        # errortype/-msg for image column
+        res = img_t.select(error=img_t.img.errortype).collect().to_pandas()
+        assert res['error'].isna().all()
+        res = img_t.select(error=img_t.img.errormsg).collect().to_pandas()
+        assert res['error'].isna().all()
+
+        for c in [t.c1, t.c1n, t.c2, t.c3, t.c4, t.c5, t.c6, t.c7]:
+            # errortype/errormsg only applies to stored computed and media columns
+            with pytest.raises(excs.Error) as excinfo:
+                _ = t.select(c.errortype).show()
+            assert 'only valid for' in str(excinfo.value)
+            with pytest.raises(excs.Error) as excinfo:
+                _ = t.select(c.errormsg).show()
+            assert 'only valid for' in str(excinfo.value)
+
+            # fileurl/localpath only applies to media columns
+            with pytest.raises(excs.Error) as excinfo:
+                _ = t.select(t.c1.fileurl).show()
+            assert 'only valid for' in str(excinfo.value)
+            with pytest.raises(excs.Error) as excinfo:
+                _ = t.select(t.c1.localpath).show()
+            assert 'only valid for' in str(excinfo.value)
+
+        # fileurl/localpath doesn't apply to unstored computed img columns
+        img_t.add_column(c9=img_t.img.rotate(30))
+        with pytest.raises(excs.Error) as excinfo:
+            _ = img_t.select(img_t.c9.localpath).show()
+        assert 'computed unstored' in str(excinfo.value)
+
+    def test_null_args(self, test_client: pxt.Client) -> None:
+        # create table with two int columns
+        schema = {'c1': FloatType(nullable=True), 'c2': FloatType(nullable=True)}
+        t = test_client.create_table('test', schema)
+
+        # computed column that doesn't allow nulls
+        t.add_column(c3=lambda c1, c2: c1 + c2, type=FloatType(nullable=False))
+        # function that does allow nulls
+        @pxt.udf(return_type=FloatType(nullable=True),
+                 param_types=[FloatType(nullable=False), FloatType(nullable=True)])
+        def f(a: int, b: int) -> int:
+            if b is None:
+                return a
+            return a + b
+        t.add_column(c4=f(t.c1, t.c2))
+
+        # data that tests all combinations of nulls
+        data = [{'c1': 1.0, 'c2': 1.0}, {'c1': 1.0, 'c2': None}, {'c1': None, 'c2': 1.0}, {'c1': None, 'c2': None}]
+        status = t.insert(data, fail_on_exception=False)
+        assert status.num_rows == len(data)
+        assert status.num_excs == len(data) - 1
+        result = t.select(t.c3, t.c4).collect()
+        assert result['c3'] == [2.0, None, None, None]
+        assert result['c4'] == [2.0, 1.0, None, None]
 
     def test_arithmetic_exprs(self, test_tbl: catalog.Table) -> None:
         t = test_tbl
 
+        _ = t[t.c2, t.c6.f3, t.c2 + t.c6.f3, (t.c2 + t.c6.f3) / (t.c6.f3 + 1)].show()
+        _ = t[t.c2 + t.c2].show()
         for op1, op2 in [(t.c2, t.c2), (t.c3, t.c3)]:
             _ = t[op1 + op2].show()
             _ = t[op1 - op2].show()
@@ -103,13 +254,13 @@ class TestExprs:
             (t.c1, t.c2), (t.c1, 1), (t.c2, t.c1), (t.c2, 'a'),
             (t.c1, t.c3), (t.c1, 1.0), (t.c3, t.c1), (t.c3, 'a')
         ]:
-            with pytest.raises(exc.OperationalError):
+            with pytest.raises(excs.Error):
                 _ = t[op1 + op2]
-            with pytest.raises(exc.OperationalError):
+            with pytest.raises(excs.Error):
                 _ = t[op1 - op2]
-            with pytest.raises(exc.OperationalError):
+            with pytest.raises(excs.Error):
                 _ = t[op1 * op2]
-            with pytest.raises(exc.OperationalError):
+            with pytest.raises(excs.Error):
                 _ = t[op1 / op2]
 
         # TODO: test division; requires predicate
@@ -117,16 +268,18 @@ class TestExprs:
             _ = t[op1 + op2].show()
             _ = t[op1 - op2].show()
             _ = t[op1 * op2].show()
+            with pytest.raises(excs.Error):
+                _ = t[op1 / op2].show()
 
         for op1, op2 in [
             (t.c6.f1, t.c6.f2), (t.c6.f1, t.c6.f3), (t.c6.f1, 1), (t.c6.f1, 1.0),
             (t.c6.f2, t.c6.f1), (t.c6.f3, t.c6.f1), (t.c6.f2, 'a'), (t.c6.f3, 'a'),
         ]:
-            with pytest.raises(exc.OperationalError):
+            with pytest.raises(excs.Error):
                 _ = t[op1 + op2].show()
-            with pytest.raises(exc.OperationalError):
+            with pytest.raises(excs.Error):
                 _ = t[op1 - op2].show()
-            with pytest.raises(exc.OperationalError):
+            with pytest.raises(excs.Error):
                 _ = t[op1 * op2].show()
 
 
@@ -138,8 +291,8 @@ class TestExprs:
 
     def test_inline_array(self, test_tbl: catalog.Table) -> None:
         t = test_tbl
-        result = t[[ [[t.c2, 1], [t.c2, 2]] ]].show()
-        t = result.col_types[0]
+        result = t.select([[t.c2, 1], [t.c2, 2]]).show()
+        t = result.column_types()[0]
         assert t.is_array_type()
         assert isinstance(t, ArrayType)
         assert t.shape == (2, 2)
@@ -167,41 +320,164 @@ class TestExprs:
         _ = t[t.c6.f1]
         _ = _.show()
         print(_)
+        # predicate on dict field
+        _ = t[t.c6.f2 < 2].show()
         #_ = t[t.c6.f2].show()
         #_ = t[t.c6.f5].show()
         _ = t[t.c6.f6.f8].show()
-        _ = t[cast(t.c6.f6.f8, ArrayType((4,), ColumnType.Type.FLOAT))].show()
+        _ = t[cast(t.c6.f6.f8, ArrayType((4,), FloatType()))].show()
 
         # top-level is array
         #_ = t[t.c7['*'].f1].show()
         #_ = t[t.c7['*'].f2].show()
         #_ = t[t.c7['*'].f5].show()
         _ = t[t.c7['*'].f6.f8].show()
-        _ = t[cast(t.c7['*'].f6.f8, ArrayType((2, 4), ColumnType.Type.FLOAT))].show()
+        _ = t[t.c7[0].f6.f8].show()
+        _ = t[t.c7[:2].f6.f8].show()
+        _ = t[t.c7[::-1].f6.f8].show()
+        _ = t[cast(t.c7['*'].f6.f8, ArrayType((2, 4), FloatType()))].show()
         print(_)
 
     def test_arrays(self, test_tbl: catalog.Table) -> None:
         t = test_tbl
-        t.add_column(catalog.Column('array_col', computed_with=[[t.c2, 1], [1, t.c2]]))
+        t.add_column(array_col=[[t.c2, 1], [1, t.c2]])
         _ = t[t.array_col].show()
         print(_)
         _ = t[t.array_col[:, 0]].show()
         print(_)
 
+    def test_astype(self, test_tbl: catalog.Table) -> None:
+        t = test_tbl
+        # Convert int to float
+        status = t.add_column(c2_as_float=t.c2.astype(FloatType()))
+        assert status.num_excs == 0
+        data = t.select(t.c2, t.c2_as_float).collect()
+        for row in data:
+            assert isinstance(row['c2'], int)
+            assert isinstance(row['c2_as_float'], float)
+            assert row['c2'] == row['c2_as_float']
+        # Compound expression
+        status = t.add_column(compound_as_float=(t.c2 + 1).astype(FloatType()))
+        assert status.num_excs == 0
+        data = t.select(t.c2, t.compound_as_float).collect()
+        for row in data:
+            assert isinstance(row['compound_as_float'], float)
+            assert row['c2'] + 1 == row['compound_as_float']
+        # Type conversion error
+        status = t.add_column(c2_as_string=t.c2.astype(StringType()))
+        assert status.num_excs == t.count()
+
+    def test_apply(self, test_tbl: catalog.Table) -> None:
+
+        t = test_tbl
+
+        # For each column c1, ..., c5, we create a new column ci_as_str that converts it to
+        # a string, then check that each row is correctly converted
+        # (For c1 this is the no-op string-to-string conversion)
+        for col_id in range(1, 6):
+            col_name = f'c{col_id}'
+            str_col_name = f'c{col_id}_str'
+            status = t.add_column(**{str_col_name: t[col_name].apply(str)})
+            assert status.num_excs == 0
+            data = t.select(t[col_name], t[str_col_name]).collect()
+            for row in data:
+                assert row[str_col_name] == str(row[col_name])
+
+        # Test a compound expression with apply
+        status = t.add_column(c2_plus_1_str=(t.c2 + 1).apply(str))
+        assert status.num_excs == 0
+        data = t.select(t.c2, t.c2_plus_1_str).collect()
+        for row in data:
+            assert row['c2_plus_1_str'] == str(row['c2'] + 1)
+
+        # For columns c6, c7, try using json.dumps and json.loads to emit and parse JSON <-> str
+        for col_id in range(6, 8):
+            col_name = f'c{col_id}'
+            str_col_name = f'c{col_id}_str'
+            back_to_json_col_name = f'c{col_id}_back_to_json'
+            status = t.add_column(**{str_col_name: t[col_name].apply(json.dumps)})
+            assert status.num_excs == 0
+            status = t.add_column(**{back_to_json_col_name: t[str_col_name].apply(json.loads)})
+            assert status.num_excs == 0
+            data = t.select(t[col_name], t[str_col_name], t[back_to_json_col_name]).collect()
+            for row in data:
+                assert row[str_col_name] == json.dumps(row[col_name])
+                assert row[back_to_json_col_name] == row[col_name]
+
+        def f1(x):
+            return str(x)
+
+        # Now test that a function without a return type throws an exception ...
+        with pytest.raises(excs.Error) as exc_info:
+            t.c2.apply(f1)
+        assert 'Column type of `f1` cannot be inferred.' in str(exc_info.value)
+
+        # ... but works if the type is specified explicitly.
+        status = t.add_column(c2_str_f1=t.c2.apply(f1, col_type=StringType()))
+        assert status.num_excs == 0
+
+        # Test that the return type of a function can be successfully inferred.
+        def f2(x) -> str:
+            return str(x)
+
+        status = t.add_column(c2_str_f2=t.c2.apply(f2))
+        assert status.num_excs == 0
+
+        # Test various validation failures.
+
+        def f3(x, y) -> str:
+            return f'{x}{y}'
+
+        with pytest.raises(excs.Error) as exc_info:
+            t.c2.apply(f3)  # Too many required parameters
+        assert str(exc_info.value) == 'Function `f3` has multiple required parameters.'
+
+        def f4() -> str:
+            return "pixeltable"
+
+        with pytest.raises(excs.Error) as exc_info:
+            t.c2.apply(f4)  # No positional parameters
+        assert str(exc_info.value) == 'Function `f4` has no positional parameters.'
+
+        def f5(**kwargs) -> str:
+            return ""
+
+        with pytest.raises(excs.Error) as exc_info:
+            t.c2.apply(f5)  # No positional parameters
+        assert str(exc_info.value) == 'Function `f5` has no positional parameters.'
+
+        # Ensure these varargs signatures are acceptable
+
+        def f6(x, **kwargs) -> str:
+            return x
+
+        t.c2.apply(f6)
+
+        def f7(x, *args) -> str:
+            return x
+
+        t.c2.apply(f7)
+
+        def f8(*args) -> str:
+            return ''
+
+        t.c2.apply(f8)
+
     def test_select_list(self, img_tbl) -> None:
         t = img_tbl
         result = t[t.img].show(n=100)
         _ = result._repr_html_()
-        df = t[t.img, udf_call(lambda img: img.rotate(60), ImageType(), tbl=t)]
-        _ = df.show(n=100)._repr_html_()
         df = t[[t.img, t.img.rotate(60)]]
         _ = df.show(n=100)._repr_html_()
 
-        with pytest.raises(exc.OperationalError):
+        with pytest.raises(excs.Error):
             _ = t[t.img.rotate]
 
     def test_img_members(self, img_tbl) -> None:
         t = img_tbl
+        # make sure the limit is applied in Python, not in the SELECT
+        result = t[t.img.height > 200][t.img].show(n=3)
+        assert len(result) == 3
         result = t[t.img.crop((10, 10, 60, 60))].show(n=100)
         result = t[t.img.crop((10, 10, 60, 60)).resize((100, 100))].show(n=100)
         result = t[t.img.crop((10, 10, 60, 60)).resize((100, 100)).convert('L')].show(n=100)
@@ -210,20 +486,21 @@ class TestExprs:
         _ = result._repr_html_()
 
     def test_img_functions(self, img_tbl) -> None:
+        skip_test_if_not_installed('nos')
         t = img_tbl
+        from pixeltable.functions.pil.image import resize
+        result = t[t.img.resize((224, 224))].show(0)
+        result = t[resize(t.img, (224, 224))].show(0)
         result = t[blend(t.img, t.img.rotate(90), 0.5)].show(100)
         print(result)
-        result = t[encode_image(t.img)].show(10)
+        from pixeltable.functions.nos.image_embedding import openai_clip
+        result = t[openai_clip(t.img.resize((224, 224)))].show(10)
         print(result)
         _ = result._repr_html_()
         _ = t.img.entropy() > 1
-        _ = _.extract_sql_predicate()
         _ = (t.img.entropy() > 1) & (t.split == 'train')
-        _ = _.extract_sql_predicate()
         _ = (t.img.entropy() > 1) & (t.split == 'train') & (t.split == 'val')
-        _ = _.extract_sql_predicate()
         _ = (t.split == 'train') & (t.img.entropy() > 1) & (t.split == 'val') & (t.img.entropy() < 0)
-        _ = _.extract_sql_predicate()
         _ = t[(t.split == 'train') & (t.category == 'n03445777')][t.img].show()
         print(_)
         result = t[t.img.width > 1].show()
@@ -235,32 +512,33 @@ class TestExprs:
         ][t.img, t.split].show()
         print(result)
 
-    def test_categoricals_map(self, img_tbl) -> None:
-        t = img_tbl
-        m = t[t.category].categorical_map()
-        _ = t[dict_map(t.category, m)].show()
-        print(_)
-
     def test_similarity(self, indexed_img_tbl: catalog.Table) -> None:
+        skip_test_if_not_installed('nos')
         t = indexed_img_tbl
         _ = t.show(30)
-        probe = t[t.img, t.category].show(1)
+        probe = t.select(t.img, t.category).show(1)
         img = probe[0, 0]
-        result = t[t.img.nearest(img)].show(10)
+        result = t.where(t.img.nearest(img)).show(10)
         assert len(result) == 10
         # nearest() with one SQL predicate and one Python predicate
         result = t[t.img.nearest(img) & (t.category == probe[0, 1]) & (t.img.width > 1)].show(10)
         # TODO: figure out how to verify results
-        #assert len(result) == 3
 
-        result = t[t.img.matches('musical instrument')].show(10)
+        with pytest.raises(excs.Error) as exc_info:
+            _ = t[t.img.nearest(img)].order_by(t.category).show()
+        assert 'cannot be used in conjunction with' in str(exc_info.value)
+
+        result = t[t.img.nearest('musical instrument')].show(10)
         assert len(result) == 10
         # matches() with one SQL predicate and one Python predicate
         french_horn_category = 'n03394916'
         result = t[
-            t.img.matches('musical instrument') & (t.category == french_horn_category) & (t.img.width > 1)
+            t.img.nearest('musical instrument') & (t.category == french_horn_category) & (t.img.width > 1)
         ].show(10)
-        #assert len(result) == 6
+
+        with pytest.raises(excs.Error) as exc_info:
+            _ = t[t.img.nearest(5)].show()
+        assert 'requires' in str(exc_info.value)
 
     # TODO: this doesn't work when combined with test_similarity(), for some reason the data table for img_tbl
     # doesn't get created; why?
@@ -269,48 +547,47 @@ class TestExprs:
         probe = t[t.img].show(1)
         img = probe[0, 0]
 
-        with pytest.raises(exc.OperationalError):
+        with pytest.raises(excs.Error):
             _ = t[t.img.nearest(img)].show(10)
-        with pytest.raises(exc.OperationalError):
-            _ = t[t.img.matches('musical instrument')].show(10)
-
-    def test_serialization(self, test_tbl: catalog.Table, img_tbl: catalog.Table) -> None:
-        t = test_tbl
-        # add array column
-        t.add_column(catalog.Column('c8', computed_with=[[1, 2, 3], [4, 5, 6]]))
-        img_t = img_tbl
-        test_exprs = [
-            t.c1,
-            t.c7['*'].f1,
-            Literal('test'),
-            InlineDict({
-                'a': t.c1, 'b': t.c6.f1, 'c': 17,
-                'd': InlineDict({'e': t.c2}),
-                'f': InlineArray((t.c3, t.c3))
-            }),
-            InlineArray([[t.c2, t.c2], [t.c2, t.c2]]),
-            t.c2 > 5,
-            ~(t.c2 > 5),
-            (t.c2 > 5) & (t.c1 == 'test'),
-            (t.c2 > 5) | (t.c1 == 'test'),
-            t.c7['*'].f5 >> [R[3], R[2], R[1], R[0]],
-            t.c8[0, 1:],
-            utils.sum_uda(t.c2).window(partition_by=t.c4, order_by=t.c3),
-        ]
-        for e in test_exprs:
+        with pytest.raises(excs.Error):
+            _ = t[t.img.nearest('musical instrument')].show(10)
+
+    def test_ids(
+            self, test_tbl: catalog.Table, test_tbl_exprs: List[exprs.Expr],
+            img_tbl: catalog.Table, img_tbl_exprs: List[exprs.Expr]
+    ) -> None:
+        d: Dict[int, exprs.Expr] = {}
+        for e in test_tbl_exprs:
+            assert e.id is not None
+            d[e.id] = e
+        for e in img_tbl_exprs:
+            assert e.id is not None
+            d[e.id] = e
+        assert len(d) == len(test_tbl_exprs) + len(img_tbl_exprs)
+
+    def test_serialization(
+            self, test_tbl_exprs: List[exprs.Expr], img_tbl_exprs: List[exprs.Expr]
+    ) -> None:
+        """Test as_dict()/from_dict() (via serialize()/deserialize()) for all exprs."""
+        for e in test_tbl_exprs:
             e_serialized = e.serialize()
-            e_deserialized = Expr.deserialize(e_serialized, t)
+            e_deserialized = Expr.deserialize(e_serialized)
             assert e.equals(e_deserialized)
 
-        img_test_exprs = [
-            img_t.img.width,
-            img_t.img.rotate(90),
-        ]
-        for e in img_test_exprs:
+        for e in img_tbl_exprs:
             e_serialized = e.serialize()
-            e_deserialized = Expr.deserialize(e_serialized, img_t)
+            e_deserialized = Expr.deserialize(e_serialized)
             assert e.equals(e_deserialized)
 
+    def test_print(self, test_tbl_exprs: List[exprs.Expr], img_tbl_exprs: List[exprs.Expr]) -> None:
+        _ = func.FunctionRegistry.get().module_fns
+        for e in test_tbl_exprs:
+            _ = str(e)
+            print(_)
+        for e in img_tbl_exprs:
+            _ = str(e)
+            print(_)
+
     def test_subexprs(self, img_tbl: catalog.Table) -> None:
         t = img_tbl
         e = t.img
@@ -318,31 +595,210 @@ class TestExprs:
         assert len(subexprs) == 1
         e = t.img.rotate(90).resize((224, 224))
         subexprs = [s for s in e.subexprs()]
-        assert len(subexprs) == 3
-        subexprs = [s for s in e.subexprs() if isinstance(s, ColumnRef)]
+        assert len(subexprs) == 4
+        subexprs = [s for s in e.subexprs(expr_class=ColumnRef)]
         assert len(subexprs) == 1
         assert t.img.equals(subexprs[0])
 
-    def test_window_fns(self, test_db: catalog.Db, test_tbl: catalog.Table) -> None:
-        db = test_db
+    def test_window_fns(self, test_client: pxt.Client, test_tbl: catalog.Table) -> None:
+        cl = test_client
         t = test_tbl
-        _ = t[utils.sum_uda(t.c2).window(partition_by=t.c4, order_by=t.c3)].show(100)
-        print(_)
+        _ = t.select(sum(t.c2, group_by=t.c4, order_by=t.c3)).show(100)
+
+        # conflicting ordering requirements
+        with pytest.raises(excs.Error):
+            _ = t.select(sum(t.c2, group_by=t.c4, order_by=t.c3), sum(t.c2, group_by=t.c3, order_by=t.c4)).show(100)
+        with pytest.raises(excs.Error):
+            _ = t.select(sum(t.c2, group_by=t.c4, order_by=t.c3), sum(t.c2, group_by=t.c3, order_by=t.c4)).show(100)
+
         # backfill works
-        t.add_column(catalog.Column('c9', computed_with=utils.sum_uda(t.c2).window(partition_by=t.c4, order_by=t.c3)))
-
-        c2 = catalog.Column('c2', IntType(), nullable=False)
-        c3 = catalog.Column('c3', FloatType(), nullable=False)
-        c4 = catalog.Column('c4', BoolType(), nullable=False)
-        new_t = db.create_table('insert_test', [c2, c3, c4])
-        new_t.add_column(catalog.Column(
-            'c2_sum', computed_with=utils.sum_uda(new_t.c2).window(partition_by=new_t.c4, order_by=new_t.c3)))
-        data_df = t[t.c2, t.c4, t.c3].show(0).to_pandas()
-        new_t.insert_pandas(data_df)
+        t.add_column(c9=sum(t.c2, group_by=t.c4, order_by=t.c3))
+        _ = t.c9.col.has_window_fn_call()
+
+        # ordering conflict between frame extraction and window fn
+        base_t = cl.create_table('videos', {'video': VideoType(), 'c2': IntType(nullable=False)})
+        args = {'video': base_t.video, 'fps': 0}
+        v = cl.create_view('frame_view', base_t, iterator_class=FrameIterator, iterator_args=args)
+        # compatible ordering
+        _ = v.select(v.frame, sum(v.frame_idx, group_by=base_t, order_by=v.pos)).show(100)
+        with pytest.raises(excs.Error):
+            # incompatible ordering
+            _ = v.select(v.frame, sum(v.c2, order_by=base_t, group_by=v.pos)).show(100)
+
+        schema = {
+            'c2': IntType(nullable=False),
+            'c3': FloatType(nullable=False),
+            'c4': BoolType(nullable=False),
+        }
+        new_t = cl.create_table('insert_test', schema=schema)
+        new_t.add_column(c2_sum=sum(new_t.c2, group_by=new_t.c4, order_by=new_t.c3))
+        rows = list(t.select(t.c2, t.c4, t.c3).collect())
+        new_t.insert(rows)
         _ = new_t.show(0)
-        print(_)
 
     def test_aggregates(self, test_tbl: catalog.Table) -> None:
         t = test_tbl
-        _ = t[t.c2 % 2, sum(t.c2), count(t.c2), sum(t.c2) + count(t.c2), sum(t.c2) + t.c2 % 2].group_by(t.c2 % 2).show()
-        print(_)
+        _ = t[t.c2 % 2, sum(t.c2), count(t.c2), sum(t.c2) + count(t.c2), sum(t.c2) + (t.c2 % 2)]\
+            .group_by(t.c2 % 2).show()
+
+        # check that aggregates don't show up in the wrong places
+        with pytest.raises(excs.Error):
+            # aggregate in where clause
+            _ = t[sum(t.c2) > 0][sum(t.c2)].group_by(t.c2 % 2).show()
+        with pytest.raises(excs.Error):
+            # aggregate in group_by clause
+            _ = t[sum(t.c2)].group_by(sum(t.c2)).show()
+        with pytest.raises(excs.Error):
+            # mixing aggregates and non-aggregates
+            _ = t[sum(t.c2) + t.c2].group_by(t.c2 % 2).show()
+        with pytest.raises(excs.Error):
+            # nested aggregates
+            _ = t[sum(count(t.c2))].group_by(t.c2 % 2).show()
+
+    def test_udas(self, test_tbl: catalog.Table) -> None:
+        t = test_tbl
+
+        @pxt.uda(
+            name='window_agg', init_types=[IntType()], update_types=[IntType()], value_type=IntType(),
+            allows_window=True, requires_order_by=False)
+        class WindowAgg:
+            def __init__(self, val: int = 0):
+                self.val = val
+            def update(self, ignore: int) -> None:
+                pass
+            def value(self) -> int:
+                return self.val
+
+        @pxt.uda(
+            name='ordered_agg', init_types=[IntType()], update_types=[IntType()], value_type=IntType(),
+            requires_order_by=True, allows_window=True)
+        class WindowAgg:
+            def __init__(self, val: int = 0):
+                self.val = val
+            def update(self, i: int) -> None:
+                pass
+            def value(self) -> int:
+                return self.val
+
+        @pxt.uda(
+            name='std_agg', init_types=[IntType()], update_types=[IntType()], value_type=IntType(),
+            requires_order_by=False, allows_window=False)
+        class StdAgg:
+            def __init__(self, val: int = 0):
+                self.val = val
+            def update(self, i: int) -> None:
+                pass
+            def value(self) -> int:
+                return self.val
+
+        # init arg is passed along
+        assert t.select(out=window_agg(t.c2, order_by=t.c2)).collect()[0]['out'] == 0
+        assert t.select(out=window_agg(t.c2, val=1, order_by=t.c2)).collect()[0]['out'] == 1
+
+        with pytest.raises(excs.Error) as exc_info:
+            _ = t.select(window_agg(t.c2, val=t.c2, order_by=t.c2)).collect()
+        assert 'needs to be a constant' in str(exc_info.value)
+
+        with pytest.raises(excs.Error) as exc_info:
+            # ordering expression not a pixeltable expr
+            _ = t.select(ordered_agg(1, t.c2)).collect()
+        assert 'but instead is a' in str(exc_info.value).lower()
+
+        with pytest.raises(excs.Error) as exc_info:
+            # explicit order_by
+            _ = t.select(ordered_agg(t.c2, order_by=t.c2)).collect()
+        assert 'order_by invalid' in str(exc_info.value).lower()
+
+        with pytest.raises(excs.Error) as exc_info:
+            # order_by for non-window function
+            _ = t.select(std_agg(t.c2, order_by=t.c2)).collect()
+        assert 'does not allow windows' in str(exc_info.value).lower()
+
+        with pytest.raises(excs.Error) as exc_info:
+            # group_by for non-window function
+            _ = t.select(std_agg(t.c2, group_by=t.c4)).collect()
+        assert 'group_by invalid' in str(exc_info.value).lower()
+
+        with pytest.raises(excs.Error) as exc_info:
+            # missing init type
+            @pxt.uda(update_types=[IntType()], value_type=IntType())
+            class WindowAgg:
+                def __init__(self, val: int = 0):
+                    self.val = val
+                def update(self, ignore: int) -> None:
+                    pass
+                def value(self) -> int:
+                    return self.val
+        assert 'init_types must be a list of' in str(exc_info.value)
+
+        with pytest.raises(excs.Error) as exc_info:
+            # missing update parameter
+            @pxt.uda(init_types=[IntType()], update_types=[], value_type=IntType())
+            class WindowAgg:
+                def __init__(self, val: int = 0):
+                    self.val = val
+                def update(self) -> None:
+                    pass
+                def value(self) -> int:
+                    return self.val
+        assert 'must have at least one parameter' in str(exc_info.value)
+
+        with pytest.raises(excs.Error) as exc_info:
+            # missing update type
+            @pxt.uda(init_types=[IntType()], update_types=[IntType()], value_type=IntType())
+            class WindowAgg:
+                def __init__(self, val: int = 0):
+                    self.val = val
+                def update(self, i1: int, i2: int) -> None:
+                    pass
+                def value(self) -> int:
+                    return self.val
+        assert 'update_types must be a list of' in str(exc_info.value)
+
+        with pytest.raises(excs.Error) as exc_info:
+            # duplicate parameter names
+            @pxt.uda(init_types=[IntType()], update_types=[IntType()], value_type=IntType())
+            class WindowAgg:
+                def __init__(self, val: int = 0):
+                    self.val = val
+                def update(self, val: int) -> None:
+                    pass
+                def value(self) -> int:
+                    return self.val
+        assert 'cannot have parameters with the same name: val' in str(exc_info.value)
+
+        with pytest.raises(excs.Error) as exc_info:
+            # invalid name
+            @pxt.uda(name='not an identifier', init_types=[IntType()], update_types=[IntType()], value_type=IntType())
+            class WindowAgg:
+                def __init__(self, val: int = 0):
+                    self.val = val
+                def update(self, i1: int, i2: int) -> None:
+                    pass
+                def value(self) -> int:
+                    return self.val
+        assert 'invalid name' in str(exc_info.value).lower()
+
+        with pytest.raises(excs.Error) as exc_info:
+            # reserved parameter name
+            @pxt.uda(init_types=[IntType()], update_types=[IntType()], value_type=IntType())
+            class WindowAgg:
+                def __init__(self, val: int = 0):
+                    self.val = val
+                def update(self, order_by: int) -> None:
+                    pass
+                def value(self) -> int:
+                    return self.val
+        assert 'order_by is reserved' in str(exc_info.value).lower()
+
+        with pytest.raises(excs.Error) as exc_info:
+            # reserved parameter name
+            @pxt.uda(init_types=[IntType()], update_types=[IntType()], value_type=IntType())
+            class WindowAgg:
+                def __init__(self, val: int = 0):
+                    self.val = val
+                def update(self, group_by: int) -> None:
+                    pass
+                def value(self) -> int:
+                    return self.val
+        assert 'group_by is reserved' in str(exc_info.value).lower()