Skip to content

numberoverzero/jsonquery

Repository files navigation

jsonquery

https://travis-ci.org/numberoverzero/jsonquery.svg?branch=master https://coveralls.io/repos/numberoverzero/jsonquery/badge.png?branch=master

Basic json -> sqlalchemy query builder

Installation

pip install jsonquery

Basic Usage

Let's define a model and get an engine set up:

from sqlalchemy import Column, Integer, String, create_engine
from sqlalchemy.orm import sessionmaker
from sqlalchemy.ext.declarative import declarative_base

Base = declarative_base()

class User(Base):
    __tablename__ = 'users'
    id = Column(Integer, primary_key=True)
    name = Column(String)
    email = Column(String)
    age = Column(Integer)
    height = Column(Integer)
engine = create_engine("sqlite://", echo=True)
Base.metadata.create_all(engine)
model = User
session = sessionmaker(bind=engine)()

We want to get all users whose name starts with 'Pat' and are at least 21:

from jsonquery import jsonquery

json = {
    "operator": "and",
    "value": [
        {
            "operator": ">=",
            "column": "age",
            "value": 21
        },
        {
            "operator": "ilike",
            "column": "name",
            "value": "pat%"
        }
    ]
}

query = jsonquery(session, User, json)
users = query.all()

Supported Data Types

jsonquery doesn't care about column type. Instead, it uses a whitelist of operators, where keys are strings (the same that would be passed in the "operator" field of a node) and the values are functions that take a column object and a value and return a sqlalchemy criterion. Here are some examples:

def greater_than(column, value):
    return column > value
register_operator(">", greater_than)

def like(column, value):
    like_func = getattr(column, 'like')
    return like_func(value)
register_operator("like", like)

By default, the following are registered:

>, >=, ==, !=, <=, <
like, ilike, in_

Use unregister_operator(opstring) to remove an operator.

Future Goals

There are a few features I want to add, but these are mostly convenience and aren't necessary to the core application, which I believe is satisfied.

Compressed and/or format

Reduce repetitive column and operator specification when possible by allowing non-scalar values for column operators. By flipping the nesting restriction on logical operators, we can omit fields specified at the column level. This is especially prominent in string matching, when the column and operator are the same, but we want to compare against 3+ values.

Currently:

{
    "operator": "or",
    "value": [
        {
            "column": "age",
            "operator": "<=",
            "value": 16
        },
        {
            "column": "age",
            "operator": ">=",
            "value": 21
        },
        {
            "column": "age",
            "operator": "==",
            "value": 18
        }
    ]
}

With compressed logical operators:

{
    "column": "age"
    "value": {
        "operator": "or",
        "value": [
            {
                "operator": "<=",
                "value": 16
            },
            {
                "operator": ">=",
                "value": 21
            },
            {
                "operator": "==",
                "value": 18
            }
        ]
    }
}

Or, when the operator is the same:

{
    "column": "name"
    "operator": "like"
    "value": {
        "operator": "or",
        "value": [
            "Bill",
            "Mary",
            "Steve"
        ]
    }
}

Contributors

About

Basic json -> sqlalchemy query builder

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages