mo-sql-parsing 8.2.21318

More SQL Parsing! Parse SQL into JSON parse tree

More SQL Parsing!

Parse SQL into JSON parse trees so we can translate it for non-sql datastores!

Branch	Status
master
dev

Problem Statement

SQL is a familiar language used to access databases. Although, each database vendor has its quirky implementation, there is enough standardization that the average developer does not need to know of those quirks. This familiar core SQL (lowest common denominator, if you will) is useful enough to explore data in primitive ways. It is hoped that, once programmers have reviewed a datastore with basic SQL queries, and they see the value of that data, and they will be motivated to use the datastore's native query format.

Objectives

The objective is to convert SQL queries to JSON-izable parse trees. This originally targeted MySQL, but has grown to include other database engines. Please paste some SQL into a new issue if it does not work for you

Project Status

November 2021 - There are over 800 tests. This parser is good enough for basic usage, including:

• inner queries,
• with clauses,
• window functions
• create/drop tables and views
• insert/update/delete statements
• lambda (->) functions

Install

pip install mo-sql-parsing

Parsing SQL

>>> from mo_sql_parsing import parse
>>> parse("select count(1) from jobs")
{'select': {'value': {'count': 1}}, 'from': 'jobs'}

Each SQL query is parsed to an object: Each clause is assigned to an object property of the same name.

>>> parse("select a as hello, b as world from jobs")
{'select': [{'value': 'a', 'name': 'hello'}, {'value': 'b', 'name': 'world'}], 'from': 'jobs'}

The SELECT clause is an array of objects containing name and value properties.

SQL Flavours

There are a few parsing modes you may be interested in:

NULL is None

The default output for this parser is to emit a null function {"null":{}} wherever NULL is encountered in the SQL. If you would like something different, you can replace nulls with None (or anything else for that matter):

result = parse(sql, null=None)

this has been implemented with a post-parse rewriting of the parse tree.

Normalized function call form (experimental)

The default behaviour of the parser is to output function calls in simple_op format: The operator being a key in the object; {op: params}. This form can be difficult to work with because the object must be scanned for known operators, or possible optional arguments, or at least distinguished from the complex query object.

You can have the parser emit function calls in normal_op format

>>> from mo_sql_parsing import parse, normal_op
>>> parse("select trim(' ' from b+c)", calls=normal_op)

which produces calls in a normalized format

{"op": op, "args": args, "kwargs": kwargs}

here is the pretty-printed JSON from the example above:

{'select': {'value': {
    'op': 'trim', 
    'args': [{'op': 'add', 'args': ['b', 'c']}], 
    'kwargs': {'characters': {'literal': ' '}}
}}}

MySQL literal strings

MySQL uses both double quotes and single quotes to declare literal strings. This is not ansi behaviour, but it is more forgiving for programmers coming from other languages. A specific parse function is provided:

result = parse_mysql(sql)

Generating SQL

You may also generate SQL from the a given JSON document. This is done by the formatter, which is in Alpha state (Oct2021).

>>> from mo_sql_parsing import format
>>> format({"from":"test", "select":["a.b", "c"]})
'SELECT a.b, c FROM test'

Contributing

In the event that the parser is not working for you, you can help make this better but simply pasting your sql (or JSON) into a new issue. Extra points if you describe the problem. Even more points if you submit a PR with a test. If you also submit a fix, then you also have my gratitude.

Run Tests

See the tests directory for instructions running tests, or writing new ones.

More about implementation

SQL queries are translated to JSON objects: Each clause is assigned to an object property of the same name.

# SELECT * FROM dual WHERE a>b ORDER BY a+b
{
    "select": "*", 
    "from": "dual", 
    "where": {"gt": ["a", "b"]}, 
    "orderby": {"value": {"add": ["a", "b"]}}
}

Expressions are also objects, but with only one property: The name of the operation, and the value holding (an array of) parameters for that operation.

{op: parameters}

and you can see this pattern in the previous example:

{"gt": ["a","b"]}

Array Programming

The mo-sql-parsing.scrub() method is used liberally throughout the code, and it "simplifies" the JSON. You may find this form a bit tedious to work with because the JSON property values can be values, lists of values, or missing. Please consider converting everything to arrays:

def listwrap(value):
    if value is None:
        return []
    elif isinstance(value, list)
        return value
    else:
        return [value]

then you may avoid all the is-it-a-list checks :

for select in listwrap(parsed_result.get('select')):
    do_something(select)

Version Changes

Version 8

• Version 8 brings flatter create table structures.
  The option list in column definition has been flattened:
  Old column format

    {"create table": {
        "columns": {
            "name": "name",
            "type": {"decimal": [2, 3]},
            "option": "not null",
        }
    }}
  

  New column format

    {"create table": {
        "columns": {
            "name": "name", 
            "type": {"decimal": [2, 3]}
            "nullable": False, 
        }
    }}
  

• Added basic DML (insert/update/delete)