sadpandajoe commented on code in PR #41803: URL: https://github.com/apache/superset/pull/41803#discussion_r3708288671
########## superset/sql/dialects/trino.py: ########## @@ -0,0 +1,369 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +from __future__ import annotations + +import typing as t + +from sqlglot import exp +from sqlglot.dialects.trino import Trino as SqlglotTrino +from sqlglot.tokens import Token, TokenType + +# Keywords that open a block terminated by ``END`` in Trino SQL routines +# (https://trino.io/docs/current/udf/sql.html). ``CASE`` is included because +# both the ``CASE`` statement and the ``CASE`` expression are terminated by +# ``END``, so counting them keeps the depth balanced either way. +BLOCK_OPENERS: set[str] = {"BEGIN", "CASE", "IF", "LOOP", "REPEAT", "WHILE"} + +# Keywords that are also scalar functions in Trino (e.g. ``IF(a, b, c)`` and +# ``REPEAT('a', 3)``). When immediately followed by ``(`` they are function +# calls, not block openers, unless the token stream shows otherwise (see +# ``_is_paren_condition_block``). +AMBIGUOUS_OPENERS: set[str] = {"IF", "REPEAT"} + +BODY_KEYWORDS: tuple[str, str] = ("RETURN", "BEGIN") + +# ``BEGIN``, ``CASE``, and ``END`` are reserved words in sqlglot's Trino +# tokenizer, so they always carry one of these dedicated token types when +# used as keywords, and a different one (``STRING``/``IDENTIFIER``) when +# used as a string literal or quoted identifier, e.g. the string ``'END'`` +# or the quoted identifier ``"end"``. ``IF``, ``LOOP``, ``REPEAT``, and +# ``WHILE`` are not reserved, so the tokenizer emits ``VAR`` for them both +# when they're used as a keyword and when they're an unquoted identifier; +# requiring ``VAR`` still rules out string literals and quoted identifiers, +# which is the ambiguity ``_is_routine_keyword`` guards against. +_RESERVED_BLOCK_TOKEN_TYPES: dict[str, TokenType] = { + "BEGIN": TokenType.BEGIN, + "CASE": TokenType.CASE, + "END": TokenType.END, +} + + +def _is_routine_keyword(token: Token, text: str) -> bool: + """ + Determine whether ``token`` (whose upper-cased text is ``text``) is an + actual occurrence of a routine block keyword, as opposed to a string + literal or quoted identifier that happens to spell the same word. + """ + if (expected := _RESERVED_BLOCK_TOKEN_TYPES.get(text)) is not None: + return token.token_type == expected + return token.token_type == TokenType.VAR + + +def _is_paren_condition_block(tokens: t.Sequence[Token], paren_index: int) -> bool: + """ + Determine whether the parenthesized group starting at ``tokens[paren_index]`` + (an ``L_PAREN``) is a procedural block condition, e.g. ``IF (a > b) THEN``, + as opposed to a scalar function call argument list, e.g. ``IF(a, b, c)``. + + Only ``IF`` has this ambiguity: a parenthesized condition is followed by + ``THEN``, while a scalar function call's closing paren never is. + """ + depth = 0 + for i in range(paren_index, len(tokens)): + token_type = tokens[i].token_type + if token_type == TokenType.L_PAREN: + depth += 1 + elif token_type == TokenType.R_PAREN: + depth -= 1 + if depth == 0: + next_token = tokens[i + 1] if i + 1 < len(tokens) else None + return ( + next_token is not None and next_token.token_type == TokenType.THEN + ) + return False + + +class InlineUDF(exp.CTE): + """ + An inline SQL user-defined function declared in a ``WITH`` clause. + + Trino supports declaring UDFs inline as part of a query:: + + WITH FUNCTION meaning_of_life() + RETURNS tinyint + BEGIN + DECLARE a tinyint DEFAULT CAST(6 AS tinyint); + DECLARE b tinyint DEFAULT CAST(7 AS tinyint); + RETURN a * b; + END + SELECT meaning_of_life() + + The function definition is stored verbatim as an opaque string (wrapped + in an ``exp.Var`` so that AST traversal helpers see an expression), since + sqlglot has no representation for SQL routine bodies. Trino does not + allow queries inside SQL UDF bodies, so no table references are hidden + by the opaque representation. + + This subclasses ``exp.CTE`` because ``sqlglot.parser.Parser._parse_with`` + only collects ``exp.CTE`` instances into the ``WITH`` clause. + """ + + arg_types = {"this": True} + + +class Trino(SqlglotTrino): + """ + Custom Trino dialect with support for inline SQL UDFs. + + sqlglot cannot parse Trino SQL routine syntax; see + https://github.com/tobymao/sqlglot/issues/5178. There are two separate + problems: + + 1. The parser splits statements on every semicolon, including the ones + inside a ``BEGIN ... END`` routine body. + 2. The ``FUNCTION`` specification in a ``WITH`` clause is not valid CTE + syntax. + + This dialect keeps routine bodies intact when splitting statements, and + parses inline function specifications into opaque `InlineUDF` nodes that + regenerate verbatim. + + Note that sqlglot's ``Dialect`` metaclass registers subclasses by class + name, so once this module is imported this class also replaces the + built-in dialect for string-based lookups (``dialect="trino"``). This is + intentional, and consistent with how other Superset dialects (e.g. + ``Dremio``) shadow their sqlglot counterparts: the extensions are purely + additive, only activating on syntax that fails to parse upstream. + """ + + class Parser(SqlglotTrino.Parser): + @staticmethod + def _block_depth_delta( + tokens: list[Token], + index: int, + prev_text: str, + ) -> int: + """ + Compute the block nesting change contributed by the routine token + at ``tokens[index]``. + """ + token = tokens[index] + text = token.text.upper() + if text in BLOCK_OPENERS: + if not _is_routine_keyword(token, text): + return 0 # string literal or quoted identifier, not a keyword + if prev_text == "END": + return 0 # block terminator, e.g. `END IF`, `END CASE` + next_token = tokens[index + 1] if index + 1 < len(tokens) else None + if ( + text in AMBIGUOUS_OPENERS + and next_token + and next_token.token_type == TokenType.L_PAREN + ): + if text == "IF" and _is_paren_condition_block(tokens, index + 1): + return 1 # procedural `IF (...) THEN`, not a call + return 0 # scalar function call, e.g. `IF(a, b, c)` + return 1 + if text == "END" and _is_routine_keyword(token, text): + return -1 + return 0 + + @staticmethod + def _starts_routine( + heads: list[TokenType], + paren_depth: int, + ) -> bool: + """ + Determine whether a ``FUNCTION`` token at the end of ``heads`` + (excluded from the list) begins a new routine specification: + ``CREATE FUNCTION``, ``CREATE OR REPLACE FUNCTION``, or an entry + in a ``WITH`` list, either right after ``WITH`` itself or after a + top-level comma separating it from a preceding CTE, e.g. + ``WITH cte AS (...), FUNCTION f() ...``. + """ + if heads[:1] == [TokenType.CREATE]: Review Comment: This preserves a block-bodied `CREATE FUNCTION` as one statement, but sqlglot represents it as `Command("CREATE")` while `SQLStatement.is_mutating()` recognizes structured `Create` (and generic `ALTER`) only, so `allow_dml=False` still lets persistent catalog-function creation reach Trino. Could we classify these `CREATE [OR REPLACE] FUNCTION` commands as mutating and add a regression through the read-only gate? ########## superset/sql/dialects/trino.py: ########## @@ -0,0 +1,369 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +from __future__ import annotations + +import typing as t + +from sqlglot import exp +from sqlglot.dialects.trino import Trino as SqlglotTrino +from sqlglot.tokens import Token, TokenType + +# Keywords that open a block terminated by ``END`` in Trino SQL routines +# (https://trino.io/docs/current/udf/sql.html). ``CASE`` is included because +# both the ``CASE`` statement and the ``CASE`` expression are terminated by +# ``END``, so counting them keeps the depth balanced either way. +BLOCK_OPENERS: set[str] = {"BEGIN", "CASE", "IF", "LOOP", "REPEAT", "WHILE"} + +# Keywords that are also scalar functions in Trino (e.g. ``IF(a, b, c)`` and +# ``REPEAT('a', 3)``). When immediately followed by ``(`` they are function +# calls, not block openers, unless the token stream shows otherwise (see +# ``_is_paren_condition_block``). +AMBIGUOUS_OPENERS: set[str] = {"IF", "REPEAT"} + +BODY_KEYWORDS: tuple[str, str] = ("RETURN", "BEGIN") + +# ``BEGIN``, ``CASE``, and ``END`` are reserved words in sqlglot's Trino +# tokenizer, so they always carry one of these dedicated token types when +# used as keywords, and a different one (``STRING``/``IDENTIFIER``) when +# used as a string literal or quoted identifier, e.g. the string ``'END'`` +# or the quoted identifier ``"end"``. ``IF``, ``LOOP``, ``REPEAT``, and +# ``WHILE`` are not reserved, so the tokenizer emits ``VAR`` for them both +# when they're used as a keyword and when they're an unquoted identifier; +# requiring ``VAR`` still rules out string literals and quoted identifiers, +# which is the ambiguity ``_is_routine_keyword`` guards against. +_RESERVED_BLOCK_TOKEN_TYPES: dict[str, TokenType] = { + "BEGIN": TokenType.BEGIN, + "CASE": TokenType.CASE, + "END": TokenType.END, +} + + +def _is_routine_keyword(token: Token, text: str) -> bool: + """ + Determine whether ``token`` (whose upper-cased text is ``text``) is an + actual occurrence of a routine block keyword, as opposed to a string + literal or quoted identifier that happens to spell the same word. + """ + if (expected := _RESERVED_BLOCK_TOKEN_TYPES.get(text)) is not None: + return token.token_type == expected + return token.token_type == TokenType.VAR + + +def _is_paren_condition_block(tokens: t.Sequence[Token], paren_index: int) -> bool: + """ + Determine whether the parenthesized group starting at ``tokens[paren_index]`` + (an ``L_PAREN``) is a procedural block condition, e.g. ``IF (a > b) THEN``, + as opposed to a scalar function call argument list, e.g. ``IF(a, b, c)``. + + Only ``IF`` has this ambiguity: a parenthesized condition is followed by + ``THEN``, while a scalar function call's closing paren never is. + """ + depth = 0 + for i in range(paren_index, len(tokens)): + token_type = tokens[i].token_type Review Comment: Trino tokenizes both these procedural words and valid unquoted identifiers as `VAR`, so a UDF parameter such as `loop` in `RETURN loop` is counted as another block opener and the real `END` leaves the body unbalanced. Could block detection use syntactic position rather than accepting every matching `VAR`, with a parameter-reference regression case? ########## superset/sql/dialects/trino.py: ########## @@ -0,0 +1,369 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +from __future__ import annotations + +import typing as t + +from sqlglot import exp +from sqlglot.dialects.trino import Trino as SqlglotTrino +from sqlglot.tokens import Token, TokenType + +# Keywords that open a block terminated by ``END`` in Trino SQL routines +# (https://trino.io/docs/current/udf/sql.html). ``CASE`` is included because +# both the ``CASE`` statement and the ``CASE`` expression are terminated by +# ``END``, so counting them keeps the depth balanced either way. +BLOCK_OPENERS: set[str] = {"BEGIN", "CASE", "IF", "LOOP", "REPEAT", "WHILE"} + +# Keywords that are also scalar functions in Trino (e.g. ``IF(a, b, c)`` and +# ``REPEAT('a', 3)``). When immediately followed by ``(`` they are function +# calls, not block openers, unless the token stream shows otherwise (see +# ``_is_paren_condition_block``). +AMBIGUOUS_OPENERS: set[str] = {"IF", "REPEAT"} + +BODY_KEYWORDS: tuple[str, str] = ("RETURN", "BEGIN") + +# ``BEGIN``, ``CASE``, and ``END`` are reserved words in sqlglot's Trino +# tokenizer, so they always carry one of these dedicated token types when +# used as keywords, and a different one (``STRING``/``IDENTIFIER``) when +# used as a string literal or quoted identifier, e.g. the string ``'END'`` +# or the quoted identifier ``"end"``. ``IF``, ``LOOP``, ``REPEAT``, and +# ``WHILE`` are not reserved, so the tokenizer emits ``VAR`` for them both +# when they're used as a keyword and when they're an unquoted identifier; +# requiring ``VAR`` still rules out string literals and quoted identifiers, +# which is the ambiguity ``_is_routine_keyword`` guards against. +_RESERVED_BLOCK_TOKEN_TYPES: dict[str, TokenType] = { + "BEGIN": TokenType.BEGIN, + "CASE": TokenType.CASE, + "END": TokenType.END, +} + + +def _is_routine_keyword(token: Token, text: str) -> bool: + """ + Determine whether ``token`` (whose upper-cased text is ``text``) is an + actual occurrence of a routine block keyword, as opposed to a string + literal or quoted identifier that happens to spell the same word. + """ + if (expected := _RESERVED_BLOCK_TOKEN_TYPES.get(text)) is not None: + return token.token_type == expected + return token.token_type == TokenType.VAR + + +def _is_paren_condition_block(tokens: t.Sequence[Token], paren_index: int) -> bool: + """ + Determine whether the parenthesized group starting at ``tokens[paren_index]`` + (an ``L_PAREN``) is a procedural block condition, e.g. ``IF (a > b) THEN``, + as opposed to a scalar function call argument list, e.g. ``IF(a, b, c)``. + + Only ``IF`` has this ambiguity: a parenthesized condition is followed by + ``THEN``, while a scalar function call's closing paren never is. + """ + depth = 0 + for i in range(paren_index, len(tokens)): + token_type = tokens[i].token_type + if token_type == TokenType.L_PAREN: + depth += 1 + elif token_type == TokenType.R_PAREN: + depth -= 1 + if depth == 0: + next_token = tokens[i + 1] if i + 1 < len(tokens) else None + return ( + next_token is not None and next_token.token_type == TokenType.THEN + ) + return False + + +class InlineUDF(exp.CTE): + """ + An inline SQL user-defined function declared in a ``WITH`` clause. + + Trino supports declaring UDFs inline as part of a query:: + + WITH FUNCTION meaning_of_life() + RETURNS tinyint + BEGIN + DECLARE a tinyint DEFAULT CAST(6 AS tinyint); + DECLARE b tinyint DEFAULT CAST(7 AS tinyint); + RETURN a * b; + END + SELECT meaning_of_life() + + The function definition is stored verbatim as an opaque string (wrapped + in an ``exp.Var`` so that AST traversal helpers see an expression), since + sqlglot has no representation for SQL routine bodies. Trino does not + allow queries inside SQL UDF bodies, so no table references are hidden + by the opaque representation. + + This subclasses ``exp.CTE`` because ``sqlglot.parser.Parser._parse_with`` + only collects ``exp.CTE`` instances into the ``WITH`` clause. + """ + + arg_types = {"this": True} + + +class Trino(SqlglotTrino): + """ + Custom Trino dialect with support for inline SQL UDFs. + + sqlglot cannot parse Trino SQL routine syntax; see + https://github.com/tobymao/sqlglot/issues/5178. There are two separate + problems: + + 1. The parser splits statements on every semicolon, including the ones + inside a ``BEGIN ... END`` routine body. + 2. The ``FUNCTION`` specification in a ``WITH`` clause is not valid CTE + syntax. + + This dialect keeps routine bodies intact when splitting statements, and + parses inline function specifications into opaque `InlineUDF` nodes that + regenerate verbatim. + + Note that sqlglot's ``Dialect`` metaclass registers subclasses by class + name, so once this module is imported this class also replaces the + built-in dialect for string-based lookups (``dialect="trino"``). This is + intentional, and consistent with how other Superset dialects (e.g. + ``Dremio``) shadow their sqlglot counterparts: the extensions are purely + additive, only activating on syntax that fails to parse upstream. + """ + + class Parser(SqlglotTrino.Parser): + @staticmethod + def _block_depth_delta( + tokens: list[Token], + index: int, + prev_text: str, + ) -> int: + """ + Compute the block nesting change contributed by the routine token + at ``tokens[index]``. + """ + token = tokens[index] + text = token.text.upper() + if text in BLOCK_OPENERS: + if not _is_routine_keyword(token, text): + return 0 # string literal or quoted identifier, not a keyword + if prev_text == "END": + return 0 # block terminator, e.g. `END IF`, `END CASE` + next_token = tokens[index + 1] if index + 1 < len(tokens) else None + if ( + text in AMBIGUOUS_OPENERS + and next_token + and next_token.token_type == TokenType.L_PAREN + ): + if text == "IF" and _is_paren_condition_block(tokens, index + 1): + return 1 # procedural `IF (...) THEN`, not a call + return 0 # scalar function call, e.g. `IF(a, b, c)` + return 1 + if text == "END" and _is_routine_keyword(token, text): + return -1 + return 0 + + @staticmethod + def _starts_routine( + heads: list[TokenType], + paren_depth: int, + ) -> bool: + """ + Determine whether a ``FUNCTION`` token at the end of ``heads`` + (excluded from the list) begins a new routine specification: + ``CREATE FUNCTION``, ``CREATE OR REPLACE FUNCTION``, or an entry + in a ``WITH`` list, either right after ``WITH`` itself or after a + top-level comma separating it from a preceding CTE, e.g. + ``WITH cte AS (...), FUNCTION f() ...``. + """ + if heads[:1] == [TokenType.CREATE]: + return heads in ( + [TokenType.CREATE], + [TokenType.CREATE, TokenType.OR, TokenType.REPLACE], + ) + if heads[:1] == [TokenType.WITH]: + return paren_depth == 0 and heads[-1] in ( + TokenType.WITH, + TokenType.COMMA, + ) + return False + + def _parse( + self, + parse_method: t.Callable[..., exp.Expression | None], + raw_tokens: list[Token], + sql: str | None = None, + ) -> list[exp.Expression | None]: + """ + Split tokens into statements, keeping routine bodies intact. + + This is a copy of ``sqlglot.parser.Parser._parse`` (as of + sqlglot 30.8.0, the version pinned in ``requirements/base.txt``) + with one change: when a statement starts with ``WITH FUNCTION``, + ``CREATE FUNCTION``, or ``CREATE OR REPLACE FUNCTION``, semicolons + inside ``BEGIN ... END`` blocks do not split the statement. If + sqlglot's own ``_parse`` changes on a future upgrade, this copy + will silently drift from it and should be re-diffed against the + new version. + """ + self.reset() + self.sql = sql or "" + + total = len(raw_tokens) + chunks: list[list[Token]] = [[]] + routine_mode: bool = False + depth: int = 0 + paren_depth: int = 0 + prev_text: str = "" + + for i, token in enumerate(raw_tokens): + if token.token_type == TokenType.SEMICOLON and depth <= 0: + if token.comments: + chunks.append([token]) + if i < total - 1: + chunks.append([]) + routine_mode = False + depth = 0 + paren_depth = 0 + prev_text = "" + continue + + chunk = chunks[-1] + chunk.append(token) + + if token.token_type == TokenType.FUNCTION and not routine_mode: + heads = [tok.token_type for tok in chunk[:-1]] + routine_mode = self._starts_routine(heads, paren_depth) + elif routine_mode: + depth += self._block_depth_delta(raw_tokens, i, prev_text) + + if token.token_type == TokenType.L_PAREN: + paren_depth += 1 + elif token.token_type == TokenType.R_PAREN: + paren_depth -= 1 + + prev_text = token.text.upper() + + self._chunks = chunks + return self._parse_batch_statements( + parse_method=parse_method, + sep_first_statement=False, + ) + + def _parse_cte(self) -> exp.CTE | None: + """ + Parse a single entry in a ``WITH`` clause. + + An entry starting with the ``FUNCTION`` keyword followed by an + identifier is an inline UDF specification; anything else + (including a CTE named "function") is handled by sqlglot. + """ + if ( + self._curr + and self._curr.token_type == TokenType.FUNCTION + and self._next + and self._next.token_type + not in (TokenType.ALIAS, TokenType.L_PAREN, TokenType.COMMA) + ): + return self._parse_inline_udf() + + return super()._parse_cte() + + def _parse_inline_udf(self) -> InlineUDF: + """ + Consume an inline UDF specification and return it verbatim. + + The specification is ``FUNCTION name(params) RETURNS type`` plus + optional routine characteristics, followed by a body that is + either ``RETURN expression`` or a ``BEGIN ... END`` block. + """ + start = self._curr + self._advance() + + # scan for the start of the function body, skipping over the + # signature, return type, and routine characteristics + paren_depth: int = 0 + body: str | None = None + while self._curr: + token_type = self._curr.token_type + text = self._curr.text.upper() + if token_type == TokenType.L_PAREN: + paren_depth += 1 + elif token_type == TokenType.R_PAREN: + paren_depth -= 1 + elif paren_depth == 0 and text in BODY_KEYWORDS: Review Comment: This compares raw token text, so valid characteristics such as `COMMENT 'RETURN'` or `COMMENT 'BEGIN'` are mistaken for the body start and parsing fails before the actual `RETURN`. Could this require an actual routine-keyword token, as the depth scanner now does, and cover both comment values? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
