feat: sql to substrait

pyproject.toml

@@ -14,7 +14,8 @@ write_to = "src/substrait/_version.py"
 [project.optional-dependencies]
 extensions = ["antlr4-python3-runtime", "pyyaml"]
 gen_proto = ["protobuf == 3.20.1", "protoletariat >= 2.0.0"]
-test = ["pytest >= 7.0.0", "antlr4-python3-runtime", "pyyaml"]
+sql = ["sqloxide", "deepdiff"]
+test = ["pytest >= 7.0.0", "antlr4-python3-runtime", "pyyaml", "sqloxide", "deepdiff", "duckdb<=1.2.2", "datafusion"]
 
 [tool.pytest.ini_options]
 pythonpath = "src"

src/substrait/builders/plan.py

@@ -26,8 +26,8 @@
 
 def _merge_extensions(*objs):
     return {
-        "extension_uris": merge_extension_uris(*[b.extension_uris for b in objs]),
-        "extensions": merge_extension_declarations(*[b.extensions for b in objs]),
+        "extension_uris": merge_extension_uris(*[b.extension_uris for b in objs if b]),
+        "extensions": merge_extension_declarations(*[b.extensions for b in objs if b]),
     }
 
 
@@ -193,13 +193,15 @@ def resolve(registry: ExtensionRegistry) -> stp.Plan:
         bound_plan = plan if isinstance(plan, stp.Plan) else plan(registry)
         ns = infer_plan_schema(bound_plan)
 
-        bound_offset = resolve_expression(offset, ns, registry)
+        bound_offset = resolve_expression(offset, ns, registry) if offset else None
         bound_count = resolve_expression(count, ns, registry)
 
         rel = stalg.Rel(
             fetch=stalg.FetchRel(
                 input=bound_plan.relations[-1].root.input,
-                offset_expr=bound_offset.referred_expr[0].expression,
+                offset_expr=bound_offset.referred_expr[0].expression
+                if bound_offset
+                else None,
                 count_expr=bound_count.referred_expr[0].expression,
             )
         )

src/substrait/sql/sql_to_substrait.py

@@ -0,0 +1,339 @@
+import random
+import string
+from sqloxide import parse_sql
+from substrait.builders.extended_expression import (
+    UnboundExtendedExpression,
+    column,
+    scalar_function,
+    literal,
+    aggregate_function,
+    window_function,
+)
+from substrait.builders.plan import (
+    read_named_table,
+    project,
+    filter,
+    sort,
+    fetch,
+    set,
+    join,
+    aggregate,
+)
+from substrait.gen.proto import type_pb2 as stt
+from substrait.gen.proto import algebra_pb2 as stalg
+from substrait.extension_registry import ExtensionRegistry
+from typing import Callable
+from deepdiff import DeepDiff
+
+SchemaResolver = Callable[[str], stt.NamedStruct]
+
+function_mapping = {
+    "Plus": ("functions_arithmetic.yaml", "add"),
+    "Minus": ("functions_arithmetic.yaml", "subtract"),
+    "Gt": ("functions_comparison.yaml", "gt"),
+    "GtEq": ("functions_comparison.yaml", "gte"),
+    "Lt": ("functions_comparison.yaml", "lt"),
+    "Eq": ("functions_comparison.yaml", "equal"),
+}
+
+aggregate_function_mapping = {"SUM": ("functions_arithmetic.yaml", "sum")}
+
+window_function_mapping = {
+    "row_number": ("functions_arithmetic.yaml", "row_number"),
+}
+
+
+def compare_dicts(dict1, dict2):
+    diff = DeepDiff(dict1, dict2, exclude_regex_paths=["span"])
+    return len(diff) == 0
+
+
+def translate_expression(
+    ast: dict,
+    schema_resolver: SchemaResolver,
+    registry: ExtensionRegistry,
+    measures: list[UnboundExtendedExpression],
+    groupings: list[dict],
+    alias: str = None,
+) -> UnboundExtendedExpression:
+    assert len(ast) == 1
+    op = list(ast.keys())[0]
+
+    if groupings:
+        # This means we are parsing a projection after a grouping
+        # Loop through used groupings for an identical ast and return it rather than recalculate
+        for i, f in enumerate(groupings):
+            if compare_dicts(ast, f):
+                return column(i, alias=alias)
+
+    ast = ast[op]
+
+    if op == "Identifier":
+        return column(ast["value"], alias=alias)
+    elif op == "UnnamedExpr" or op == "expr" or op == "Unnamed" or op == "Expr":
+        return translate_expression(
+            ast,
+            schema_resolver=schema_resolver,
+            registry=registry,
+            measures=measures,
+            groupings=groupings,
+        )
+    elif op == "ExprWithAlias":
+        return translate_expression(
+            ast["expr"],
+            schema_resolver=schema_resolver,
+            registry=registry,
+            measures=measures,
+            groupings=groupings,
+            alias=ast["alias"]["value"],
+        )
+    elif op == "BinaryOp":
+        expressions = [
+            translate_expression(
+                ast["left"],
+                schema_resolver=schema_resolver,
+                registry=registry,
+                measures=measures,
+                groupings=groupings,
+            ),
+            translate_expression(
+                ast["right"],
+                schema_resolver=schema_resolver,
+                registry=registry,
+                measures=measures,
+                groupings=groupings,
+            ),
+        ]
+        func = function_mapping[ast["op"]]
+        return scalar_function(func[0], func[1], expressions=expressions, alias=alias)
+    elif op == "Value":
+        return literal(
+            int(ast["value"]["Number"][0]), stt.Type(i64=stt.Type.I64()), alias=alias
+        )  # TODO infer type
+    elif op == "Function":
+        expressions = [
+            translate_expression(
+                e,
+                schema_resolver=schema_resolver,
+                registry=registry,
+                measures=measures,
+                groupings=groupings,
+            )
+            for e in ast["args"]["List"]["args"]
+        ]
+        name = ast["name"][0]["Identifier"]["value"]
+
+        if name in function_mapping:
+            func = function_mapping[name]
+            return scalar_function(func[0], func[1], *expressions, alias=alias)
+        elif name in aggregate_function_mapping:
+            # All measures need to be extracted out because substrait calculates measures in a separate rel
+            # We generate a random name for the measure and return a column with that name for the projection to work
+            # Start by checking if multiple measures are identical and reuse previously generated name
+            for m in measures:
+                if compare_dicts(ast, m[1]):
+                    return column(m[2], alias=alias)
+
+            func = aggregate_function_mapping[name]
+            random_name = "".join(
+                random.choices(string.ascii_uppercase + string.digits, k=5)
+            )  # TODO make this deterministic
+            aggr = aggregate_function(func[0], func[1], expressions, alias=random_name)
+            measures.append((aggr, ast, random_name))
+            return column(random_name, alias=alias)
+        elif name in window_function_mapping:
+            func = window_function_mapping[name]
+
+            partitions = [
+                translate_expression(
+                    e,
+                    schema_resolver=schema_resolver,
+                    registry=registry,
+                    measures=measures,
+                    groupings=groupings,
+                )
+                for e in ast["over"]["WindowSpec"]["partition_by"]
+            ]
+
+            return window_function(
+                func[0], func[1], expressions, partitions=partitions, alias=alias
+            )
+
+        else:
+            raise Exception(f"Unknown function {name}")
+    # elif op == "Wildcard":
+    #     return wildcard()
+    else:
+        raise Exception(f"Unknown op {op}")
+
+
+def translate(ast: dict, schema_resolver: SchemaResolver, registry: ExtensionRegistry):
+    assert len(ast) == 1
+    op = list(ast.keys())[0]
+    ast = ast[op]
+
+    if op == "Query":
+        relation = translate(
+            ast["body"], schema_resolver=schema_resolver, registry=registry
+        )
+
+        if ast["order_by"]:
+            expressions = [
+                translate_expression(
+                    e["expr"],
+                    schema_resolver=schema_resolver,
+                    registry=registry,
+                    measures=None,
+                    groupings=None,
+                )
+                for e in ast["order_by"]["kind"]["Expressions"]
+            ]
+            relation = sort(relation, expressions)(registry)
+
+        if ast["limit_clause"]:
+            limit_expression = translate_expression(
+                ast["limit_clause"]["LimitOffset"]["limit"],
+                schema_resolver=schema_resolver,
+                registry=registry,
+                measures=None,
+                groupings=None,
+            )
+
+            if ast["limit_clause"]["LimitOffset"]["offset"]:
+                offset_expression = translate_expression(
+                    ast["limit_clause"]["LimitOffset"]["offset"]["value"],
+                    schema_resolver=schema_resolver,
+                    registry=registry,
+                    measures=None,
+                    groupings=None,
+                )
+            else:
+                offset_expression = None
+
+            relation = fetch(relation, offset_expression, limit_expression)(registry)
+
+        return relation
+    elif op == "Select":
+        relation = translate(
+            ast["from"][0]["relation"],
+            schema_resolver=schema_resolver,
+            registry=registry,
+        )
+
+        if ast["from"][0]["joins"]:
+            for _join in ast["from"][0]["joins"]:
+                join_type_mapping = {
+                    "Inner": stalg.JoinRel.JOIN_TYPE_INNER,
+                    "Left": stalg.JoinRel.JOIN_TYPE_LEFT,
+                    "LeftOuter": stalg.JoinRel.JOIN_TYPE_LEFT,
+                    "RightOuter": stalg.JoinRel.JOIN_TYPE_RIGHT,
+                    "Right": stalg.JoinRel.JOIN_TYPE_RIGHT,
+                }
+                right = translate(
+                    _join["relation"],
+                    schema_resolver=schema_resolver,
+                    registry=registry,
+                )
+
+                join_type = list(_join["join_operator"].keys())[0]
+
+                expression = translate_expression(
+                    _join["join_operator"][join_type]["On"],
+                    schema_resolver=schema_resolver,
+                    registry=registry,
+                    measures=None,
+                    groupings=None,
+                )
+
+                relation = join(
+                    relation, right, expression, join_type_mapping[join_type]
+                )(registry)
+
+        if "selection" in ast and ast["selection"]:
+            where_expression = translate_expression(
+                ast["selection"],
+                schema_resolver=schema_resolver,
+                registry=registry,
+                measures=None,
+                groupings=None,
+            )
+            relation = filter(relation, where_expression)(registry)
+
+        if ast["group_by"] and ast["group_by"]["Expressions"][0]:
+            groupings = ast["group_by"]["Expressions"][0]
+            grouping_expressions = [
+                translate_expression(
+                    e,
+                    schema_resolver=schema_resolver,
+                    registry=registry,
+                    measures=None,
+                    groupings=None,
+                )
+                for e in groupings
+            ]
+        else:
+            groupings = []
+            grouping_expressions = []
+
+        measures = []
+
+        projection = [
+            translate_expression(
+                p,
+                schema_resolver=schema_resolver,
+                registry=registry,
+                measures=measures,
+                groupings=groupings,
+            )
+            for p in ast["projection"]
+        ]
+
+        if ast["having"]:
+            having_predicate = translate_expression(
+                ast["having"],
+                schema_resolver=schema_resolver,
+                registry=registry,
+                measures=measures,
+                groupings=[],
+            )
+        else:
+            having_predicate = None
+
+        if measures or groupings:
+            relation = aggregate(
+                relation, grouping_expressions, [e[0] for e in measures]
+            )(registry)
+
+        if having_predicate:
+            relation = filter(relation, having_predicate)(registry)
+
+        return project(relation, expressions=projection)(registry)
+    elif op == "Table":
+        name = ast["name"][0]["Identifier"]["value"]
+        return read_named_table(name, schema_resolver(name))
+    elif op == "SetOperation":
+        # TODO more than 2 inputs to a set operation
+        left = translate(
+            ast["left"], schema_resolver=schema_resolver, registry=registry
+        )
+        right = translate(
+            ast["right"], schema_resolver=schema_resolver, registry=registry
+        )
+        if ast["op"] == "Union":
+            set_op = (
+                stalg.SetRel.SET_OP_UNION_ALL
+                if ast["set_quantifier"] == "All"
+                else stalg.SetRel.SET_OP_UNION_DISTINCT
+            )
+        else:
+            raise Exception("")
+
+        return set([left, right], set_op)(registry)
+    else:
+        raise Exception(f"Unknown op {op}")
+
+
+def convert(query: str, dialect: str, schema_resolver: SchemaResolver):
+    ast = parse_sql(sql=query, dialect=dialect)[0]
+    registry = ExtensionRegistry(load_default_extensions=True)
+    return translate(ast, schema_resolver=schema_resolver, registry=registry)