SQL Parser CST

SQL Parser which produces a Concrete Syntax Tree (CST).

Unlike a more usual parser which produces an Abstract Syntax Tree (AST), this parser preserves all the syntax elements present in the parsed source code, with the goal of being able to re-create the exact original source code.

Try it live in SQL Explorer.

Features

• Detailed TypeScript types for the syntax tree
• Unified syntax tree for multiple SQL dialects
• Includes source code location data for all nodes
• Includes comments in the syntax tree
• Helpful error messages
• Fast

Supports the following SQL dialects:

• SQLite - full support (version 3.45)
• BigQuery - full support (as of 31 January 2024).
• MySQL - experimental (version 8) (see #7 for implementation progress).
• MariaDB - experimental (version 10) (see #32 for implementation progress).
• PostgreSQL - experimental (version 16) (see #40 for implementation progress).
• PL/pgSQL - experimental (version 16) (see #130 for implementation progress).

Note: This software is in very active development. The syntax tree structure is mostly stable now, though there are bound to be changes as new SQL dialects are added and they contain features that need to be accommodated to the syntax tree.

Install

npm install sql-parser-cst

Usage

import { parse, show, cstVisitor } from "sql-parser-cst";

const cst = parse("select * from my_table", {
  dialect: "sqlite",
  // These are optional:
  includeSpaces: true, // Adds spaces/tabs
  includeNewlines: true, // Adds newlines
  includeComments: true, // Adds comments
  includeRange: true, // Adds source code location data
});

// convert all keywords to uppercase
const toUpper = cstVisitor({
  keyword: (kw) => {
    kw.text = kw.text.toUpperCase();
  },
});
toUpper(cst);

// Serialize back to SQL
show(cst); // --> SELECT * FROM my_table

AST versus CST-parsers

For example, given the following SQL:

/* My query */
SELECT ("first_name" || ' jr.') as fname
-- use important table
FROM persons;

An AST-parser might parse this to the following abstract syntax tree:

{
  "type": "select_stmt",
  "columns": [
    {
      "type": "alias",
      "expr": {
        "type": "binary_expr",
        "left": { "type": "column_ref", "column": "first_name" },
        "operator": "||",
        "right": { "type": "string", "value": " jr." }
      },
      "alias": "fname"
    }
  ],
  "from": [{ "type": "table_ref", "table": "persons" }]
}

Note that the above AST is missing the following information:

• comments
• whitespace (e.g. where the newlines are)
• case of keywords (e.g. whether AS or as was written)
• whether an identifier was quoted or not (and with what kind of quotes)
• whether an expression is wrapped in additional (unnecessary) parenthesis.
• whether the statement ends with a semicolon.

In contrast, this CST parser produces the following concrete syntax tree, which preserves all of this information:

{
  "type": "program",
  "statements": [
    {
      "type": "select_stmt",
      "clauses": [
        {
          "type": "select_clause",
          "selectKw": { "type": "keyword", "text": "SELECT", "name": "SELECT" },
          "options": [],
          "columns": {
            "type": "list_expr",
            "items": [
              {
                "type": "alias",
                "expr": {
                  "type": "paren_expr",
                  "expr": {
                    "type": "binary_expr",
                    "left": { "type": "identifier", "text": "\"first_name\"", "name": "first_name" },
                    "operator": "||",
                    "right": { "type": "string_literal", "text": "' jr.'", "value": " jr." }
                  }
                },
                "asKw": { "type": "keyword", "text": "as", "name": "AS" },
                "alias": { "type": "identifier", "text": "fname", "name": "fname" }
              }
            ]
          }
        },
        {
          "type": "from_clause",
          "fromKw": { "type": "keyword", "text": "FROM", "name": "FROM" },
          "expr": { "type": "identifier", "text": "persons", "name": "persons" },
          "leading": [
            { "type": "newline", "text": "\n" },
            { "type": "line_comment", "text": "-- use important table" },
            { "type": "newline", "text": "\n" }
          ]
        }
      ]
    },
    { "type": "empty" }
  ],
  "leading": [
    { "type": "block_comment", "text": "/* My query */" },
    { "type": "newline", "text": "\n" }
  ]
}

Note the following conventions:

• All keywords are preserved in type: keyword nodes, which are usually stored in fields named like someNameKw.
• Parenthesis is represented by separate type: paren_expr node.
• Comma-separated lists are represented by separate type: list_expr node.
• Trailing semicolon is represented by type: empty node in the end.
• The original source code representation of strings, identifiers, keywords, etc is preserved in text fields.
• Each node can have leading and trailing fields, which store comments and newlines immediately before or after that node. These fields will also contain information about regular spaces/tabs (e.g. {"type": "space", "text": " \t"}). This has been left out from this example for the sake of simplicity.

API

parse(sql: string, options: ParserOptions): Program

Parses SQL string and returns the CST tree. Takes the following options:

• dialect: 'sqlite' | 'bigquery' | 'mysql' | 'mariadb' | 'postgresql' | 'plpgsql' The SQL dialect to parse (required).
• includeRange: boolean When enabled adds range: [number, number] field to all CST nodes, which contains the start and end locations of the node.
• includeComments: boolean When enabled adds leading: Whitespace[] and/or trailing: Whitespace[] to nodes which are preceded or followed by comments.
• includeNewlines: boolean Like includeComments, but includes newlines info to the same fields.
• includeSpaces: boolean Like includeComments, but includes horizontal whitespace info to the same fields.
• paramTypes: ("?" | "?nr" | "$nr" | ":name" | "$name" | "@name" | "@`name`")[] Determines the types of bound parameters supported by the parser. By default a query like SELECT * FROM tbl WHERE id = ? will result in parse error. To fix it, use paramTypes: ["?"] config option.
• filename: string Name of the SQL file. This is only used for error-reporting.

When parsing fails with syntax error, it throws FormattedSyntaxError which contains a message like:

Syntax Error: Unexpected "WHERE"
Was expecting to see: "!", "$", "(", "-", ":", "?", "@", "CASE", ...
--> my_db.sql:2:33
  |
2 | SELECT * FROM my_table ORDER BY WHERE
  |                                 ^

Because not all SQL dialects are 100% supported and even the ones that are can have bugs, you might encounter SQL that the parser fails to parse. As a workaround you can use conditional comments to disable the parser for the problematic part of SQL:

SELECT * FROM products;
/* sql-parser-cst-disable */
UNSUPPORTED SQL IN HERE;
/* sql-parser-cst-enable */
UPDATE products SET sold = true WHERE id = 2;

show(cst: Node): string

Converts CST back to string.

Important caveat: the CST has to contain whitespace data, meaning, it was generated with includeComments, includeNewlines and includeSpaces options enabled.

For any valid SQL the following assertion will always hold:

const opts = {
  dialect: "sqlite",
  includeComments: true,
  includeNewlines: true,
  includeSpaces: true,
};

show(parse(sql, opts)) === sql; // always true

cstVisitor(map: VisitorMap): (node: Node) => SKIP | void

Generates a function that walks through the whole CST tree and calls a function in map whenever it encounters a node with that type.

For example the following code checks that all table and column aliases use the explicit AS keyword:

const checkAliases = cstVisitor({
  alias: (node) => {
    if (!node.asKw) {
      throw new Error("All alias definitions must use AS keyword!");
    }
  },
});
checkAliases(cst);

You can return VisitorAction.SKIP to avoid visiting all child nodes of a specific node:

let topLevelSelects = 0;
const countTopLevelSelects = cstVisitor({
  select_stmt: (node) => {
    topLevelSelects++;
    return VisitorAction.SKIP;
  },
});
countTopLevelSelects(cst);

cstTransformer<T>(map: TransformMap<T>): (node: Node) => T

Transforms the whole CST into some other type T. The map object should contain an entry for each of the CST node types it expects to encounter (this generally means all of them).

For example, the following implements a toString() function that serializes very basic SQL queries like SELECT 1, 2, 3 + 4:

const toString = cstTransformer({
  program: (node) => node.statements.map(toString).join(";"),
  select_statement: (node) => node.clauses.map(toString).join(" "),
  select_clause: (node) => "SELECT " + node.columns.map(toString).join(", "),
  binary_expr: (node) =>
    toString(node.left) + " " + node.operator + " " + toString(node.right),
  number_literal: (node) => node.text,
});

The builtin show() function is implemented as such a transform.

xKeywords: Record<string, boolean>

Additionally the parser exports lists of reserved keywords for each supported SQL dialect: sqliteKeywords, bigqueryKeywords, mysqlKeywords, mariadbKeywords, postgresqlKeywords. These are simple JavaScript objects, useful for doing lookups:

export const sqliteKeywords = {
  ABORT: true,
  ACTION: true,
  ADD: true,
  ...
};

Development

yarn generate will generate parser.

The testsuite contains two kinds of tests:

• tests applicable for all dialects
• tests applicable for only some specific dialects

When running the testsuite one always needs to pick a dialect. For example yarn test:sqlite or yarn test:mysql. Running one of these commands will run the testsuite against the parser of that dialect. It will execute all the generic tests plus tests applicable for that dialect.

yarn test will execute the testsuite for each supported dialect, covering all the possible combinations.

During development

Start the parser-generator watch process in one terminal:

yarn watch:generate

and the tests watch process in another terminal:

yarn test:sqlite --watch

Note that yarn test --watch doesn't work. A separate watch process needs to be started manually for each dialect.

Release

Generate new release with yarn publish.

To generate a changelog use the yarn changelog command:

VERSION=v0.27.0 yarn changelog

Acknowledgements

This started as a fork of node-sql-parser, which is based on @flora/sql-parser, which in turn was extracted from Alibaba's nquery module.

There's very little left of the original code though.