RBQL: Rainbow Query Language

RBQL is an eval-based SQL-like query engine for (not only) CSV file processing. It provides SQL-like language that supports SELECT queries with Python or JavaScript expressions.
RBQL is best suited for data transformation, data cleaning, and analytical queries.
RBQL is distributed with CLI apps, text editor plugins, Python and JS libraries.

Official Site

Supported formats

Matrix of data formats that RBQL supports out of the box. R=Read, W=Write

If you use RBQL as a library you can write implementation for a couple of classes to support additional formats.

Main Features

• Use Python or JavaScript expressions inside SELECT, UPDATE, WHERE and ORDER BY statements
• Supports multiple input formats
• Result set of any query immediately becomes a first-class table on its own
• No need to provide FROM statement in the query - input table is defined by the current context
• Supports all main SQL keywords
• Supports aggregate functions and GROUP BY queries
• Supports user-defined functions (UDF)
• Provides some new useful query modes which traditional SQL engines do not have
• Lightweight, dependency-free, works out of the box

Limitations:

• RBQL doesn't support nested queries, but they can be emulated with consecutive queries
• Number of tables in all JOIN queries is always 2 (input table and join table), use consecutive queries to join 3 or more tables

Supported SQL Keywords (Keywords are case insensitive)

• SELECT
• UPDATE
• WHERE
• ORDER BY ... [ DESC | ASC ]
• [ LEFT | INNER ] JOIN
• DISTINCT
• GROUP BY
• TOP N
• LIMIT N

All keywords have the same meaning as in SQL queries. You can check them online

RBQL variables

RBQL for CSV files provides the following variables which you can use in your queries:

• a1, a2,..., a{N}
  Variable type: string
  Description: value of i-th field in the current record in input table
• b1, b2,..., b{N}
  Variable type: string
  Description: value of i-th field in the current record in join table B
• NR
  Variable type: integer
  Description: Record number (1-based)
• NF
  Variable type: integer
  Description: Number of fields in the current record
• a.name, b.Person_age, ... a.{Good_alphanumeric_column_name}
  Variable type: string
  Description: Value of the field referenced by it's "name". You can use this notation if the field in the header has a "good" alphanumeric name
• a["object id"], a['9.12341234'], b["%$ !! 10 20"] ... a["Arbitrary column name!"]
  Variable type: string
  Description: Value of the field referenced by it's "name". You can use this notation to reference fields by arbitrary values in the header

UPDATE statement

UPDATE query produces a new table where original values are replaced according to the UPDATE expression, so it can also be considered a special type of SELECT query. This prevents accidental data loss from poorly written queries.
UPDATE SET is synonym to UPDATE, because in RBQL there is no need to specify the source table.

Aggregate functions and queries

RBQL supports the following aggregate functions, which can also be used with GROUP BY keyword:
COUNT, ARRAY_AGG, MIN, MAX, SUM, AVG, VARIANCE, MEDIAN

Limitation: aggregate functions inside Python (or JS) expressions are not supported. Although you can use expressions inside aggregate functions.
E.g. MAX(float(a1) / 1000) - valid; MAX(a1) / 1000 - invalid.
There is a workaround for the limitation above for ARRAY_AGG function which supports an optional parameter - a callback function that can do something with the aggregated array. Example:
select a2, ARRAY_AGG(a1, lambda v: sorted(v)[:5]) group by a2 - Python; select a2, ARRAY_AGG(a1, v => v.sort().slice(0, 5)) group by a2 - JS

JOIN statements

Join table B can be referenced either by its file path or by its name - an arbitrary string which the user should provide before executing the JOIN query.
RBQL supports STRICT LEFT JOIN which is like LEFT JOIN, but generates an error if any key in the left table "A" doesn't have exactly one matching key in the right table "B".
Limitation: JOIN statements can't contain Python/JS expressions and must have the following form: <JOIN_KEYWORD> (/path/to/table.tsv | table_name ) ON a... == b... [AND a... == b... [AND ... ]]

SELECT EXCEPT statement

SELECT EXCEPT can be used to select everything except specific columns. E.g. to select everything but columns 2 and 4, run: SELECT * EXCEPT a2, a4
Traditional SQL engines do not support this query mode.

UNNEST() operator

UNNEST(list) takes a list/array as an argument and repeats the output record multiple times - one time for each value from the list argument.
Example: SELECT a1, UNNEST(a2.split(';'))

LIKE() function

RBQL does not support LIKE operator, instead it provides "like()" function which can be used like this: SELECT * where like(a1, 'foo%bar')

WITH (header) and WITH (noheader) statements

You can set whether the input (and join) CSV file has a header or not using the environment configuration parameters which could be --with_headers CLI flag or GUI checkbox or something else. But it is also possible to override this selection directly in the query by adding either WITH (header) or WITH (noheader) statement at the end of the query. Example: select top 5 NR, * with (header)

User Defined Functions (UDF)

RBQL supports User Defined Functions
You can define custom functions and/or import libraries in two special files:

• ~/.rbql_init_source.py - for Python
• ~/.rbql_init_source.js - for JavaScript

Examples of RBQL queries

With Python expressions

• select top 100 a1, int(a2) * 10, len(a4) where a1 == "Buy" order by int(a2) desc
• select * order by random.random() - random sort
• select len(a.vehicle_price) / 10, a2 where int(a.vehicle_price) < 500 and a['Vehicle type'] in ["car", "plane", "boat"] limit 20 - referencing columns by names from header and using Python's "in" to emulate SQL's "in"
• update set a3 = 'NPC' where a3.find('Non-playable character') != -1
• select NR, * - enumerate records, NR is 1-based
• select * where re.match(".*ab.*", a1) is not None - select entries where first column has "ab" pattern
• select a1, b1, b2 inner join ./countries.txt on a2 == b1 order by a1, a3 - example of join query
• select MAX(a1), MIN(a1) where a.Name != 'John' group by a2, a3 - example of aggregate query
• select *a1.split(':') - Using Python3 unpack operator to split one column into many. Do not try this with other SQL engines!

With JavaScript expressions

• select top 100 a1, a2 * 10, a4.length where a1 == "Buy" order by parseInt(a2) desc
• select * order by Math.random() - random sort
• select top 20 a.vehicle_price.length / 10, a2 where parseInt(a.vehicle_price) < 500 && ["car", "plane", "boat"].indexOf(a['Vehicle type']) > -1 limit 20 - referencing columns by names from header
• update set a3 = 'NPC' where a3.indexOf('Non-playable character') != -1
• select NR, * - enumerate records, NR is 1-based
• select a1, b1, b2 inner join ./countries.txt on a2 == b1 order by a1, a3 - example of join query
• select MAX(a1), MIN(a1) where a.Name != 'John' group by a2, a3 - example of aggregate query
• select ...a1.split(':') - Using JS "destructuring assignment" syntax to split one column into many. Do not try this with other SQL engines!

RBQL design principles and architecture

RBQL core idea is based on dynamic code generation and execution with exec and eval functions. Here are the main steps that RBQL engine performs when processing a query:

1. Shallow parsing: split the query into logical expressions such as "SELECT", "WHERE", "ORDER BY", etc.
2. Embed the expression segments into the main loop template code
3. Execute the hydrated loop code

Here you can find a very basic working script (only 15 lines of Python code) which implements this idea: mini_rbql.py

The diagram below gives an overview of the main RBQL components and data flow:

Advantages of RBQL over traditional SQL engines

• Provides power and flexibility of general purpose Python and JS languages in relational expressions (including regexp, math, file system, json, xml, random and many other libraries that these languages provide)
• Can work with different data sources including CSV files, sqlite tables, native 2D arrays/lists (traditional SQL engines are usually tightly coupled with their databases)
• Result set of any query immediately becomes a first-class table on its own
• Supports both TOP and LIMIT keywords
• Provides additional NR (record number) variable which is especially useful for input sources where record order is well defined (such as CSV files)
• Supports input tables with inconsistent number of fields per record
• Allows to generate result sets with variable number of fields per record e.g. by using split() function and unpack operator (Python) / destructuring assignment (JS)
• UPDATE is a special case of SELECT query - this prevents accidental data loss
• No need to use FROM statement - the table name is defined by the context. This improves query typing speed and allows immediate autocomplete for variables inside SELECT statement (in traditional SQL engines autocomplete will not work until you write FROM statement, which goes after SELECT statement)
• SELECT, WHERE, ORDER BY, and other statements can be rearranged in any way you like
• Supports EXCEPT statement
• Provides a fully-functional client-side browser demo application
• Almost nonexistent entry barrier both for SQL users and JS/Python users
• Integration with popular text editors (VSCode, Vim, Sublime Text, Atom)
• Small, maintainable, dependency-free, eco-friendly and hackable code base: RBQL engine fits into a single file with less than 2000 LOC

Disadvantages of RBQL compared to traditional SQL engines

• Not suitable for transactional workload
• RBQL doesn't support nested queries, but they can be emulated with consecutive queries
• Number of tables in all JOIN queries is always 2 (input table and join table), use consecutive queries to join 3 or more tables
• Does not support HAVING statement

References

• RBQL: Official Site
• RBQL is integrated with Rainbow CSV extensions in Vim, VSCode, Sublime Text and Atom editors.
• RBQL in npm: $ npm install -g rbql
• RBQL in PyPI: $ pip install rbql