KSON: JSON with SQL and Networking

Of course it's a good idea: what makes you ask?

KSON is a superset of JSON with the following features:

• Remote document references (so you can embed a JSON, KSON, or CSV file available at a public URL or file path)
• Embedded SQL: Write queries against other objects in your JSON file (including references and deeply nested objects) with the full power of SQLite and have the queries evaluate to JSON
• Use comments (/* ... */) and global named references ("foo": "bar" as myRef).
• Compiles to JSON: Run kson file.kson (see installation instructions below) and boom! you have JSON.

KSON combines the portability of the top data exchange formats (JSON, CSV) with the expressiveness of the leading data querying language (SQL), and the flexibility of dynamic embedded references. This project is pre-release and we welcome bug reports and any other feedback.

Installation

Run

python3 -m pip install ksonpy

This will create a global executable kson which you can run on .kson files to produce .json output:

kson file.kson [--indent <integer>]

or pipe to a file:

kson file.kson > file.json

Examples

You can find examples in the examples/ directory.

• examples/gdp.kson: Demonstrates how you can query an external data source (in this case, CSV file on GitHub.)
• examples/join-gdp-and-population.kson: Fetch data from two data sources ( GDP by country and population by country) and perform a join to see GDP per capita.
• examples/nested-references.kson: Often external JSON data buries the important data in a nested structure. We can dereference arbitrarily deep into it.

Grammar and Semantics

All of the JSON grammar is supported. In addition:

• A reference is denoted by <<url>>, where url is any non-empty string of characters that can be requested over the network or the local filesystem. By default, a reference will be compiled to its full contents in the generated JSON. To avoid that, you can write a reference like this: <<!url>>, with an exclamation point. Now the reference will compile to to just url.
  • By default, a reference will attempt to automatically discover whether it's formatted as a JSON, CSV, or KSON, falling back to a string constant. You can provide a type hint like so: <<url|json>>, <<url|csv>>, etc.
  • If a URL cannot be resolved, an error will be thrown and JSON generation will fail.
  • To be precise, a remote JSON file will be compiled to its exact contents; a remote CSV file will be compiled to a list of dictionaries, where a best-effort attempt is made to convert numeric values to integers and/or floats; and a KSON file will be just treated as more KSON (although the namespace of aliases must remain globally unique, as discussed below.)
• An alias can be added after any token - a string, an array, a dictionary, and especially after a reference. Suppose you have a token token (perhaps token is "hello world", or 123, or <<https://json.org>>) -- then you may also write the token as token as myAliasName, eg 123 as myNumber, <<https://json.org>> as someRef.
  • Aliases must be globally unique (as a consequence, it would be a syntax error for a kson file with aliases to reference itself).
  • The alias can be referenced in SQL queries by prepending with $; for example, an alias someRef can be addressed as $someRef.
  • However, not all aliases will be pointing at something useful for a SQL query; we can only write queries against scalar values (strings and numbers) or against tables. A table can be constructed from a list of dictionaries pointing to scalars with consistent types. We can also coerce a list of scalars to a table by constructing a table with a single column whose name is the alias for the list.
  • We will recurse on nested dictionary structures until we find a scalar or table value. For instance, suppose that you reference a remote document as in Fig. 1, aliased as doc, and you wish to query the list of scalars baz. Then in your SQL query, you can call select * from "$doc$foo$bar$baz". (As an aside, quotes are generally required around the SQL alias references to parse properly.)
  • You must define an alias before it is used in a SQL query, so typically references will be placed toward the top of a file.

{
  "foo": {
    "bar": {
      "baz": [
        1,
        2,
        3,
        4
      ]
    },
    "someValue": 42
  }
}

Fig. 1

• A SQL query is delimited by triple backticks (```) before and after. SQL queries can contain aliases to objects in your document, as described above. The SQL queries are executed by an in-memory sqlite engine.
  • The output of a SQL query is usually returned as a list of dictionaries, but there are two exceptions. If only one column is in the output, a list of scalars will be returned instead. If there is only one column and, in addition, you use the directive limit 1 in your query, then the result will be returned as a single scalar value.

Tests

You can run the tests by cloning the repository and then:

python3 -m pip install -r requirements.txt
python3 -m unittest tests

In all candidness, there are likely many edge cases, so if you find one, it would be greatly appreciated if you would make a bug report and we can add a regression test.

FAQ

How does this work?

It's pretty simple, actually: First we parse the KSON file. Where JSON has arrays and dictionaries, we throw in a few extra types - refs, aliases, and SQL queries.

To compile the file, we traverse the tree, making network requests, building appropriately-named SQLite tables, and performing SQL queries as we go, eventually collapsing the whole business to JSON.

Some constraints of this approach are that we make network requests in serial, and that you must define an alias before any SQL queries which use it.

What's the motivation for this project?

For reasons which are best elided, I had to write an enormous number of JSON parsers in a short period of time, and then got some additional ideas about the format. It's called "KSON" because k comes after j, get it? :-)