tree-sitter-sql-polyglot

SQL grammars for tree-sitter: a strict ANSI base and 22 dialects built on top of it. Each dialect is its own parser, so Postgres code is parsed as Postgres and T-SQL as T-SQL.

Forked from DerekStride/tree-sitter-sql, which mixes several dialects into one permissive grammar.

Install

pip install tree-sitter-sql-polyglot
npm install @relativelyunknown/tree-sitter-sql-polyglot
cargo add tree-sitter-sql-polyglot --features postgres   # or --features full for all 22
go get github.com/relativelyunknown/tree-sitter-sql-polyglot/bindings/go/postgres

Swift (SwiftPM) and C (CMake) are supported too. See the usage docs.

from tree_sitter import Language, Parser
import tree_sitter_sql

parser = Parser(Language(tree_sitter_sql.language_postgres()))
tree = parser.parse(b"SELECT id FROM users WHERE name ILIKE 'a%'")

import Parser from "tree-sitter";
import { postgres } from "@relativelyunknown/tree-sitter-sql-polyglot";

const parser = new Parser();
parser.setLanguage(postgres);
const tree = parser.parse("SELECT id FROM users WHERE name ILIKE 'a%'");

Only the dialects you use get compiled or loaded. Each one also ships its own highlights query (HIGHLIGHTS_QUERY_POSTGRES in Python and Rust, postgres.HIGHLIGHTS_QUERY in Node).

Dialects

Dialect	Extends	Notable syntax
base		ANSI SQL: GROUPING SETS, OFFSET ... FETCH, WITHIN GROUP, GRANT/REVOKE
postgres	base	COPY, VACUUM, :: casts, PARTITION OF, row-level security policies
cockroachdb	postgres	AS OF SYSTEM TIME, UPSERT, BACKUP/RESTORE, changefeeds
mysql	base	ENGINE=, index hints, SHOW, LIMIT offset, count, @@ variables
mariadb	mysql	system-versioned tables, RETURNING, INVISIBLE columns
sqlite	base	INSERT OR REPLACE, AUTOINCREMENT, INDEXED BY
oracle	base	PL/SQL blocks and packages, CONNECT BY, BULK COLLECT
db2	base	SQL PL, modules, audit policies, federated objects
tsql	base	T-SQL scripting, CROSS APPLY, query hints, #temp tables
hana	base	column/row tables, UPSERT ... WITH PRIMARY KEY, SQLScript
teradata	base	SEL/DEL, PRIMARY INDEX, RANGE_N, COLLECT STATISTICS
bigquery	base	STRUCT<...>/ARRAY<...>, UNNEST, QUALIFY
spanner	bigquery	INTERLEAVE IN PARENT, change streams, row deletion policies
snowflake	base	scripting, LATERAL FLATTEN, time travel, @stage
redshift	base	DISTKEY/, external schemas, /

A child dialect only adds what its engine adds; it gets the rest from its parent. How much of each engine's syntax is covered, checked against SQLGlot, ANTLR, pglast and sqlfluff, is on the coverage page.

Development

You need Node and the tree-sitter CLI (npm install -g tree-sitter-cli).

npm install                  # unpacks the committed parsers and builds the Node addons
npm run generate             # regenerate the base parser after editing grammar/
npm run generate -- postgres # or a single dialect
npm run generate:all         # or everything
npm run test:corpus          # base corpus tests
(cd postgres && npx tree-sitter test)  # a dialect's corpus tests

A base grammar change affects every dialect, so regenerate and test all of them. Generation is cached by file hash; npm run generate:force skips the cache.

CONTRIBUTING.md covers the workflow and AGENTS.md the grammar layout.

Related

• DerekStride/tree-sitter-sql, the upstream grammar
• takegue/tree-sitter-sql-bigquery
• m-novikov/tree-sitter-sql

License

MIT. The git history from upstream is kept, so upstream authors show up in the contributors list; CODEOWNERS lists who maintains this fork. LICENSE carries both copyright notices.