grai-io / sqlglot

Python SQL Parser and Transpiler

SQLGlot

SQLGlot is a no dependency Python SQL parser, transpiler, and optimizer. It can be used to format SQL or translate between different dialects like Presto, Spark, and Hive. It aims to read a wide variety of SQL inputs and output syntatically correct SQL in the targeted dialects.

It is currently the fastest pure-Python SQL parser.

You can easily customize the parser to support UDF's across dialects as well through the transform API.

Syntax errors are highlighted and dialect incompatibilities can warn or raise depending on configurations.

Install

From PyPI

pip3 install sqlglot

Or with a local checkout

pip3 install -e .

Examples

Easily translate from one dialect to another. For example, date/time functions vary from dialects and can be hard to deal with.

import sqlglot
sqlglot.transpile("SELECT EPOCH_MS(1618088028295)", read='duckdb', write='hive')

SELECT TO_UTC_TIMESTAMP(FROM_UNIXTIME(1618088028295 / 1000, 'yyyy-MM-dd HH:mm:ss'), 'UTC')

SQLGlot can even translate custom time formats.

import sqlglot
sqlglot.transpile("SELECT STRFTIME(x, '%y-%-m-%S')", read='duckdb', write='hive')

SELECT DATE_FORMAT(x, 'yy-M-ss')"

Formatting and Transpiling

Read in a SQL statement with a CTE and CASTING to a REAL and then transpiling to Spark.

Spark uses backticks as identifiers and the REAL type is transpiled to FLOAT.

import sqlglot

sql = """WITH baz AS (SELECT a, c FROM foo WHERE a = 1) SELECT f.a, b.b, baz.c, CAST("b"."a" AS REAL) d FROM foo f JOIN bar b ON f.a = b.a LEFT JOIN baz ON f.a = baz.a"""
sqlglot.transpile(sql, write='spark', identify=True, pretty=True)[0]

WITH `baz` AS (
  SELECT
    `a`,
    `c`
  FROM `foo`
  WHERE
    `a` = 1
)
SELECT
  `f`.`a`,
  `b`.`b`,
  `baz`.`c`,
  CAST(`b`.`a` AS FLOAT) AS `d`
FROM `foo` AS `f`
JOIN `bar` AS `b`
  ON `f`.`a` = `b`.`a`
LEFT JOIN `baz`
  ON `f`.`a` = `baz`.`a`

Metadata

You can explore SQL with expression helpers to do things like find columns and tables.

from sqlglot import parse_one, exp

# print all column references (a and b)
for column in parse_one("SELECT a, b + 1 AS c FROM d").find_all(exp.Column):
  print(column.alias_or_name)

# find all projections in select statements (a and c)
for select in parse_one("SELECT a, b + 1 AS c FROM d").find_all(exp.Select):
  for projection in select.args["expressions"]:
    print(projection.alias_or_name)

# find all tables (x, y, z)
for table in parse_one("SELECT * FROM x JOIN y JOIN z").find_all(exp.Table):
  print(table.name)

Parser Errors

A syntax error will result in a parser error.

transpile("SELECT foo( FROM bar")

sqlglot.errors.ParseError: Expecting ). Line 1, Col: 13.
  SELECT foo( FROM bar

Unsupported Errors

Presto APPROX_DISTINCT supports the accuracy argument which is not supported in Spark.

transpile(
    'SELECT APPROX_DISTINCT(a, 0.1) FROM foo',
    read='presto',
    write='spark',
)

WARNING:root:APPROX_COUNT_DISTINCT does not support accuracy

SELECT APPROX_COUNT_DISTINCT(a) FROM foo

Build and Modify SQL

SQLGlot supports incrementally building sql expressions.

from sqlglot import select, condition

where = condition("x=1").and_("y=1")
select("*").from_("y").where(where).sql()

Which outputs:

SELECT * FROM y WHERE x = 1 AND y = 1

You can also modify a parsed tree:

from sqlglot import parse_one

parse_one("SELECT x FROM y").from_("z").sql()

Which outputs:

SELECT x FROM y, z

There is also a way to recursively transform the parsed tree by applying a mapping function to each tree node:

from sqlglot import exp, parse_one

expression_tree = parse_one("SELECT a FROM x")

def transformer(node):
    if isinstance(node, exp.Column) and node.name == "a":
        return parse_one("FUN(a)")
    return node

transformed_tree = expression_tree.transform(transformer)
transformed_tree.sql()

Which outputs:

SELECT FUN(a) FROM x

SQL Optimizer

SQLGlot can rewrite queries into an "optimized" form. It performs a variety of techniques to create a new canonical AST. This AST can be used to standaradize queries or provide the foundations for implementing an actual engine.

import sqlglot
from sqlglot.optimizer import optimize

>>>
optimize(
    sqlglot.parse_one("""
    SELECT A OR (B OR (C AND D))
    FROM x
    WHERE Z = date '2021-01-01' + INTERVAL '1' month OR 1 = 0
    """),
    schema={"x": {"A": "INT", "B": "INT", "C": "INT", "D": "INT", "Z": "STRING"}}
).sql(pretty=True)

"""
SELECT
  (
    "x"."A"
    OR "x"."B"
    OR "x"."C"
  )
  AND (
    "x"."A"
    OR "x"."B"
    OR "x"."D"
  ) AS "_col_0"
FROM "x" AS "x"
WHERE
  "x"."Z" = CAST('2021-02-01' AS DATE)
"""

SQL Annotations

SQLGlot supports annotations in the sql expression. This is an experimental feature that is not part of any of the SQL standards but it can be useful when needing to annotate what a selected field is supposed to be. Below is an example:

SELECT
  user #primary_key,
  country
FROM users

Customization

Custom Types

A simple transform on types can be accomplished by providing a corresponding mapping:

from sqlglot import *

transpile("SELECT CAST(a AS INT) FROM x", type_mapping={exp.DataType.Type.INT: "SPECIAL INT"})[0]

SELECT CAST(a AS SPECIAL INT) FROM x

More complicated transforms can be accomplished by using the Tokenizer, Parser, and Generator directly.

Custom Functions

In this example, we want to parse a UDF SPECIAL_UDF and then output another version called SPECIAL_UDF_INVERSE with the arguments switched.

from sqlglot import *
from sqlglot.expressions import Func

class SpecialUdf(Func):
    arg_types = {'a': True, 'b': True}

tokens = Tokenizer().tokenize("SELECT SPECIAL_UDF(a, b) FROM x")

Here is the output of the tokenizer:

[
    <Token token_type: TokenType.SELECT, text: SELECT, line: 0, col: 0>,
    <Token token_type: TokenType.VAR, text: SPECIAL_UDF, line: 0, col: 7>,
    <Token token_type: TokenType.L_PAREN, text: (, line: 0, col: 18>,
    <Token token_type: TokenType.VAR, text: a, line: 0, col: 19>,
    <Token token_type: TokenType.COMMA, text: ,, line: 0, col: 20>,
    <Token token_type: TokenType.VAR, text: b, line: 0, col: 22>,
    <Token token_type: TokenType.R_PAREN, text: ), line: 0, col: 23>,
    <Token token_type: TokenType.FROM, text: FROM, line: 0, col: 25>,
    <Token token_type: TokenType.VAR, text: x, line: 0, col: 30>,
]

expression = Parser(functions={
    **SpecialUdf.default_parser_mappings(),
}).parse(tokens)[0]

repr(expression)

The expression tree produced by the parser:

(SELECT distinct: False, expressions:
  (SPECIALUDF a:
    (COLUMN this:
      (IDENTIFIER this: a, quoted: False)), b:
    (COLUMN this:
      (IDENTIFIER this: b, quoted: False))), from:
  (FROM expressions:
    (TABLE this:
      (IDENTIFIER this: x, quoted: False))))

Finally generating the new SQL:

Generator(transforms={
    SpecialUdf: lambda self, e: f"SPECIAL_UDF_INVERSE({self.sql(e, 'b')}, {self.sql(e, 'a')})"
}).generate(expression)

SELECT SPECIAL_UDF_INVERSE(b, a) FROM x

Benchmarks

Benchmarks run on Python 3.9.6 in seconds.

Query	sqlglot	sqlparse	moz_sql_parser	sqloxide
short	0.00038	0.00104	0.00174	0.000060
long	0.00508	0.01522	0.02162	0.000597
crazy	0.01871	3.49415	0.35346	0.003104

Run Tests and Lint

pip install -r requirements.txt
./format_code.sh
./run_checks.sh

Optional Dependencies

SQLGlot uses dateutil to simplify literal timedelta expressions. The optimizer will not simplify expressions like

x + interval '1' month

if the module cannot be found.