Skip to content

Latest commit

 

History

History
1074 lines (789 loc) · 49.7 KB

CONTRIBUTING.md

File metadata and controls

1074 lines (789 loc) · 49.7 KB
Raw

Contributing To The CCXT Library

How To Submit An Issue

If you want to submit an issue and you want your issue to be resolved quickly, here's a checklist for you:

  • Read the Manual, and especially carefully read the following sections:
  • Read the Troubleshooting section and follow troubleshooting steps.
  • Read the FAQ for most frequently asked questions.
  • Read the API docs for your exchange.
  • Search for similar issues first to avoid duplicates.
  • If your issue is unique, along with a basic description of the failure, the following IS REQUIRED:
    • set exchange.verbose = true property on the exchange instance before calling its functions or methods
    • DON'T POST SCREENSHOTS OF CODE OR ERRORS, POST THE OUTPUT AND CODE IN PLAIN TEXT!
    • surround code and output with triple backticks: ```GOOD```
    • don't confuse the backtick symbol (`) with the quote symbol ('): '''BAD'''
    • don't confuse a single backtick with triple backticks: `BAD`
    • paste a complete code snippet you're having difficulties with, avoid one-liners
    • paste the full verbose output of the failing method without your keys
    • the verbose output should include the request and response from the exchange (not just an error callstack)
    • write your language and version
    • write ccxt library version
    • which exchange it is
    • which method you're trying to call

Reporting Vulnerabilities And Critical Issues

If you found a security issue or a critical vulnerability and reporting it in public would impose risk – please feel free to send us a message to info@ccxt.trade.

How To Contribute Code

  • MAKE SURE YOUR CODE IS UNIFIED!

    ↑ This is the most important rule of all!!!

  • PLEASE, DO NOT COMMIT THE FOLLOWING FILES IN PULL REQUESTS:

    • /doc/* (these files are generated from /wiki/*, place your edits there)
    • /build/* (these are generated automatically)
    • /php/* (except for base classes)
    • /python/* (except for base classes)
    • /ccxt.js
    • /README.md (exchange lists are generated automatically)
    • /package.json
    • /package.lock
    • /wiki/* (except for real edits, exchange lists are generated automatically)
    • /dist/ccxt.browser.js (this is also browserified automatically)

    These files are generated (explained below) and will be overwritten upon build. Please don't commit them to avoid bloating the repository which is already quite large. Most often, you have to commit just one single source file to submit an edit to the implementation of an exchange.

  • PLEASE, SUBMIT ATOMIC EDITS, ONE PULL REQUEST PER ONE EXCHANGE, DO NOT MIX THEM

  • MAKE SURE YOUR CODE PASSES ALL SYNTAX CHECKS BY RUNNING npm run build

Pending Tasks

Below is a list of functionality we would like to have implemented and fully unified in the library in the first place at this time. Most of these tasks are already in progress, implemented for some exchanges, but not all of them:

  • Margin trading
  • Leverage
  • Derivatives (futures, options)
  • Main account / subaccounts
  • Conditional orders (stop loss, take profit)
  • transfer between subaccounts and main account
  • fetchTransfer
  • fetchTransfers
  • fetchLedger
  • fetchPositions
  • closePosition
  • closePositions

If you want to contribute by submitting partial implementations be sure to look up examples of how it's done inside the library (where implemented already) and copy the adopted practices.

If your proposal, suggestion or improvement does not relate to the above list of tasks before submitting it make sure it is:

  1. really needed by the majority of ccxt users
  2. designed to be a general-purpose solution, not hardcoded for your specific needs
  3. done in a generalized way compatible with all exchanges (not exchange-specific)
  4. portable (available in all supported languages)
  5. robust
  6. explicit in what it's doing
  7. doesn't break anything (if you change a method, make sure that all other methods calling the edited method are not broken)

The following is a set of rules for contributing to the ccxt library codebase.

What You Need To Have

If you're not going to develop CCXT and contribute code to the CCXT library, then you don't need the Docker image nor the CCXT repository. If you just want to use CCXT inside your project simply install it as a regular package into the project folder as documented in the Manual (https://github.com/ccxt/ccxt/wiki/Install):

With Docker

The easiest way is to use Docker to run an isolated build & test enviroment with all the dependencies installed:

docker-compose run --rm ccxt

That builds a container and opens a shell, where the npm run build and node run-tests commands should simply work out of the box.

The CCXT folder is mapped inside of the container, except the node_modules folder — the container would have its own ephemeral copy — so that won't mess up your locally installed modules. This means that you can edit sources on your host machine using your favorite editor and build/test them in the running container.

This way you can keep the build tools and processes isolated, not having to work through the painful process of installing all those dependencies to your host machine manually.

Without Docker

Dependencies

  • Git
  • Node.js 8+
  • Python 3.5.3+
    • requests (pip install requests)
    • aiohttp (pip install aiohttp)
    • tox
      • via pip: pip install tox
      • MacOS with brew: brew install tox
      • Ubuntu Linux: apt-get install tox
  • PHP 5.3+ with the following extensions installed and enabled:
    • cURL
    • iconv
    • mbstring
    • PCRE
    • bcmath (php<7.1)

Build Steps

git clone https://github.com/ccxt/ccxt.git
cd ccxt
npm install
npm run build

What You Need To Know

Repository Structure

The contents of the repository are structured as follows:

/                          # root directory aka npm module/package folder for Node.js
/.babelrc                  # babel config used for making the ES5 version of the library
/.eslintrc                 # linter
/.gitattributes            # contains linguist settings for language detection in repo
/.gitignore                # ignore it
/.npmignore                # files to exclude from the NPM package
/.travis.yml               # a YAML config for travis-ci (continuous integration)
/CONTRIBUTING.md           # this file
/LICENSE.txt               # MIT
/README.md                 # master markdown for GitHub, npmjs.com, npms.io, yarn and others
/build/                    # build scripts
/build/export-exchanges.js # used to create tables of exchanges in the docs during the build
/build/transpile.js        # the transpilation script
/build/update-badges.js    # a JS script to update badges in the README and in docs
/build/vss.js              # reads single-sourced version from package.json and writes it everywhere
/dist/                     # a folder for the generated browser bundle of CCXT
/ccxt.js                   # entry point for the master JS version of the ccxt library
/ccxt.php                  # entry point for the PHP version of the ccxt library
/doc/                      # Sphinx-generated rst-docs for http://docs.ccxt.com/
/js/                       # the JS version of the library
/php/                      # PHP ccxt module/package folder
/python/                   # Python ccxt module/package folder for PyPI
/python/__init__.py        # entry point for the Python version of the ccxt.library
/python/async_support/     # asynchronous version of the ccxt.library for Python 3.5.3+ asyncio
/python/base/              # base code for the Python version of the ccxt library
/python/MANIFEST.in        # a PyPI-package file listing extra package files (license, configs, etc...)
/python/README.md          # a copy of README.md for PyPI
/python/setup.cfg          # wheels config file for the Python package
/python/setup.py           # pip/setuptools script (build/install) for ccxt in Python
/python/tox.ini            # tox config for Python
/examples/                 # self-explanatory
/examples/js               # ...
/examples/php              # ...
/examples/py               # ...
/exchanges.cfg             # custom bundle config for including only the exchanges you need
/package.json              # npm package file, also used in setup.py for version single-sourcing
/run-tests.js              # a front-end to run individual tests of all exchanges in all languages (JS/PHP/Python)
/wiki/                     # the source of all docs (edits go here)

Multilanguage Support

The ccxt library is available in three different languages (more to come). We encourage developers to design portable code, so that a single-language user could read the code in other languages and understand it easily. This helps the adoption of the library. The main goal is to provide a generalized, unified, consistent and robust interface to as many existing cryptocurrency exchanges as possible.

At first, all language-specific versions were developed in parallel, but separately from each other. But when it became too hard to maintain and keep the code consistent among all supported languages we have decided to switch to what we call a source/generated process. There is now a single source version in one language, that is JavaScript. Other language-specific versions are syntactically derived (transpiled, generated) automatically from the source version. But it doesn't mean that you have to be a JS coder to contribute. The portability principle allows Python and PHP devs to effectively participate in developing the source version as well.

The module entry points are:

  • ./python/__init__.py for the Python pip package
  • ./python/async/__init__.py for the Python 3.5.3+ ccxt.async_support subpackage
  • ./ccxt.js for the Node.js npm package
  • ./dist/ccxt.browser.js for the browser bundle
  • ./ccxt.php for PHP

Generated versions and docs are transpiled from the source ccxt.js file and files in ./js/ by the npm run build command.

Transpiled (generated) files

  • All derived exchange classes are transpiled automatically from source JS files. The source files are language-agnostic, easily mapped line-to-line to any other language and written in a cross-language-compatible way. Any coder can read it (by design).
  • All base classes are not transpiled, those are language-specific.

JavaScript

The ccxt.browser.js is generated with Babel from source.

Python

These files containing derived exchange classes are transpiled from JS into Python:

  • js/[_a-z].js → python/ccxt/async/[_a-z].py
  • python/ccxt/async[_a-z].py → python/ccxt/[_a-z].py (Python 3 asyncio → Python sync transpilation stage)
  • python/ccxt/test/test_async.py → python/ccxt/test/test_sync.py (the sync test is generated from the async test)

These Python base classes and files are not transpiled:

  • python/ccxt/base/*
  • python/ccxt/async/base/*

PHP

These files containing derived exchange classes are transpiled from JS into PHP:

  • js/[_a-z].js → php/[_a-z].php

These PHP base classes and files are not transpiled:

  • php/Exchange.php php/ExchangeError.php php/Precise.php ...

Typescript

  • js/[_a-z].js → ccxt.d.ts

Base Class

UNDER CONSTRUCTION

Derived Exchange Classes

Transpiler is regex-based and heavily relies on specific formatting rules. If you break them then the transpiler will either fail to generate Python/PHP classes at all or will generate malformed Python/PHP syntax.

Below are key notes on how to keep the JS code transpileable.

Use the linter npm run lint js/your-exchange-implementation.js before you build. It will cover many (but not all) the issues, so manual checking will still be required if transpilation fails.

If you see a [TypeError] Cannot read property '1' of null exception or any other transpilation error when you npm run build, check if your code satisfies the following rules:

  • don't put empty lines inside your methods
  • always use Python-style indentation, it is preserved as is for all languages
  • indent with 4 spaces exactly, avoid tabs
  • put an empty line between each of your methods
  • avoid mixed comment styles, use double-slash // in JS for line comments
  • avoid multi-line comments

If the transpiling process finishes successfully, but generates incorrect Python/PHP syntax, check for the following:

  • every opening bracket like ( or { should have a space before it!
  • do not use language-specific code syntax sugar, even if you really want to
  • unfold all maps and comprehensions to basic for-loops
  • don't change the arguments of overridden inherited methods, keep them uniform across all exchanges
  • everything should be done using base class methods only (for example, use this.json () for converting objects to json)
  • always put a semicolon ; at the end of each statement, as in PHP/C-style
  • all associative keys must be single-quoted strings everywhere (array['good']), do not use the dot notation (array.bad)
  • never use the var keyword, instead use const for constants or let for variables

And structurally:

  • if you need another base method you will have to implement it in all three languages
  • try not to issue more than one HTTP request from a unified method
  • avoid changing the contents of the arguments and params passed by reference into function calls
  • keep it simple, don't do more than one statement in one line
  • try to reduce syntax & logic (if possible) to basic one-liner expressions
  • multiple lines are ok, but you should avoid deep nesting with lots of brackets
  • do not use conditional statements that are too complex (heavy if-bracketing)
  • do not use heavy ternary conditionals
  • avoid operators clutter (don't do this: a && b || c ? d + 80 : e ** f)
  • do not use .includes(), use .indexOf() instead!
  • never use .toString() on floats: Number (0.00000001).toString () === '1e-8'
  • do not use closures, a.map or a.filter (x => (x === 'foobar')) is not acceptable in derived classes
  • do not use the in operator to check if a value is in a non-associative array (list)
  • don't add custom currency or symbol/pair conversions and formatting, copy from existing code instead
  • don't access non-existent keys, array['key'] || {} won't work in other languages!

Sending Market Ids

Most of exchanges' API endpoints will require an exchange-specific market symbol or trading pair or instrument to be specified in the request.

We don't send unified symbols to exchanges directly! They are not interchangeable! There is a significant difference between exchange-specific market-ids and unified symbols! This is explained in the Manual, here:

NEVER DO THIS:

async fetchTicker (symbol, params = {}) {
   const request = {
      'pair': symbol, // very bad, sending unified symbols to the exchange directly
   };
   const response = await this.publicGetEndpoint (request);
   // parse in a unified way...
}

DO NOT DO THIS EITHER:

async fetchTicker (symbol, params = {}) {
   const request = {
      'symbol': symbol, // very bad, sending unified symbols to the exchange directly
   };
   const response = await this.publicGetEndpoint (request);
   // parse in a unified way...
}

Instead of sending a unified CCXT symbol to the exchange, we always take the exchange-specific market-id that corresponds to that symbol. If it so happens that an exchange specific market-id is exactly the same as the CCXT unified symbol – that's a happy coincidence, but we never rely on that in the unified CCXT API.

To get the exchange-specific market-id by a unified CCXT symbol, use the following methods:

  • this.market (symbol) – returns the entire unified market structure, containing the id, baseId, quoteId, and many other interesting things
  • this.marketId (symbol) – returns just the exchange-specific id of a market by a unified symbol (if you don't need anything else)

GOOD EXAMPLES:

async fetchTicker (symbol, params = {}) {
   const market = this.market (symbol); // the entire market structure
   const request = {
      'pair': market['id'], // good, they may be equal, but often differ, it's ok
   };
   const response = await this.publicGetEndpoint (this.extend (request, params));
   // parse in a unified way...
}
async fetchTicker (symbol, params = {}) {
   const marketId = this.marketId (symbol); // just the id
   const request = {
      'symbol': marketId, // good, they may be equal, but often differ, it's ok
   };
   const response = await this.publicGetEndpoint (this.extend (request, params));
   // parse in a unified way...
}

Parsing Symbols

When sending requests to the exchange unified symbols have to be "converted" to exchange-specific market-ids like shown above. The same is true on the other end – when receiving an exchange response it has an exchange-specific market-id inside it that has to be "converted back" to a unified CCXT symbol.

We don't put exchange-specific market-ids in unified structures directly! We can't freely interchange symbols with ids! There is a significant difference between an exchange-specific market-ids and unified symbols! This is explained in the Manual, here:

NEVER DO THIS::

parseTrade (trade, market = undefined) {
   // parsing code...
   return {
      'info': trade,
      'symbol': trade['pair'], // very bad, returning exchange-specific market-ids in a unified structure!
      // other fields...
   };
}

DO NOT DO THIS EITHER

parseTrade (trade, market = undefined) {
   // parsing code...
   return {
      'info': trade,
      'symbol': trade['symbol'], // very bad, returning exchange-specific market-ids in a unified structure!
      // other fields...
   };
}

In order to handle the market-id properly it has to be looked-up in the info cached on this exchange with loadMarkets():

GOOD EXAMPLE:

parseTrade (trade, market = undefined) {
    const marketId = this.safeString (trade, 'pair');
    // safeSymbol is used to parse the market id to a unified symbol
    const symbol = this.safeSymbol (marketId, market);
    return {
       'info': trade,
       'symbol': symbol, // very good, a unified symbol here now
       // other fields...
    };
}

Accessing Dictionary Keys

In JavaScript, dictionary keys can be accessed in two notations:

  • object['key'] (single-quoted string key notation)
  • object.key (property notation)

Both work almost identically, and one is implicitly converted to another upon executing the JavaScript code.

While the above does work in JavaScript, it will not work in Python or PHP. In most languages, associative dictionary keys are not treated in the same way as properties. Therefore, in Python object.key is not the same as object['key']. In PHP $object->key is not the same as $object['key'] as well. Languages that differentiate between associative keys and properties use different notations for the two.

To keep the code transpileable, please, remember this simple rule: always use the single-quoted string key notation object['key'] for accessing all associative dictionary keys in all languages everywhere throughout this library!

Sanitizing Input With safe-Methods

JavaScript is less restrictive than other languages. It will tolerate an attempt to dereference a non-existent key where other languages will throw an Exception:

// JavaScript

const someObject = {}

if (someObject['nonExistentKey']) {
    // the body of this conditional will not execute in JavaScript
    // because someObject['nonExistentKey'] === undefined === false
    // but JavaScript will not throw an exception on the if-clause
}

However, the above logic with "an undefined value by default" will not work in Python or PHP.

# Python
some_dictionary = {}

# breaks
if some_dictionary['nonExistentKey']:
    # in Python the attempt to dereference the nonExistentKey value
    # will throw a standard built-in KeyError exception

# works
if 'nonExistentKey' in some_dictionary and some_dictionary['nonExistentKey']:
    # this is a way to check if the key exists before accessing the value

# also works
if some_dictionary.get('nonExistentKey'):
    # another a way to check if the key exists before accessing the value...

Most languages will not tolerate an attempt to access a non-existent key in an object.

For the above reasons, please, never do this in the transpiled JS files:

// JavaScript
const value = object['key'] || other_value; // will not work in Python or PHP!
if (object['key'] || other_value) { /* will not work in Python or PHP! */ }

Therefore we have a family of safe* functions:

  • safeInteger (object, key, default), safeInteger2 (object, key1, key2, default) –for parsing timestamps in milliseconds
  • safeNumber (object, key, default), safeNumber2 (object, key1, key2, default) – for parsing amounts, prices, costs
  • safeString (object, key, default), safeString2 (object, key1, key2, default) – for parsing ids, types, statuses
  • safeStringLower (object, key, default), safeStringLower2 (object, key1, key2, default) – for parsing and turning to lowercase
  • safeStringUpper (object, key, default), safeStringUpper2 (object, key1, key2, default) – for parsing and turning to lowercase
  • safeValue (object, key, default), safeValue2 (object, key1, key2, default) – for parsing objects (dictionaries) and arrays (lists)
  • safeTimestamp (object, key, default), safeTimestamp2 (object, key1, key2, default) – for parsing UNIX timestamps in seconds

The safeValue function is used for objects inside objects, arrays inside objects and boolean true/false values.

If you need to search for several different keys within an object you have available the safeMethodN function's family that allows for a search with an arbitrary number of keys by accepting an array of keys as an argument.

const price = this.safeStringN (object, [ 'key1', 'key2', 'key3' ], default)

For every safe method listed above, there is the correspondent safeMethodN too.

The above safe-functions will check for the existence of the key (or key1, key2) in the object and will properly return undefined/None/null values for JS/Python/PHP. Each function also accepts the default value to be returned instead of undefined/None/null in the last argument.

Alternatively, you could check for the key existence first...

So, you have to change this:

if (params['foo'] !== undefined) {
}

↓

const foo = this.safeValue (params, 'foo');
if (foo !== undefined) {
}

Or:

if ('foo' in params) {
}

Using Base Class Cryptography Methods For Authentication

Do not reinvent the wheel. Always use base-class methods for cryptography.

The CCXT library supports the following authentication algorithms and cryptography algorithms:

  • HMAC
  • JWT (JSON Web Token)
  • RSA
  • ECDSA Elliptic Curve Cryptography
    • NIST P256
    • secp256k1
  • OTP 2FA (one-time password 2-factor authentication)

The base Exchange class offers several methods that are key to practically all cryptography in this lib. Derived exchange implementations must not use external dependencies for cryptography, everything should be done with base methods only.

  • hash (message, hash = 'md5', digest = 'hex')
  • hmac (message, secret, hash = 'sha256', digest = 'hex')
  • jwt (message, secret, hash = 'HS256')
  • rsa (message, secret, alg = 'RS256')
  • ecdsa (request, secret, algorithm = 'p256', hash = undefined)
  • totp (secret)
  • stringToBase64(), base64ToBinary(), binaryToBase64()...

The hash() method supports the following hash algorithms:

  • 'md5'
  • 'sha1'
  • 'sha3'
  • 'sha256'
  • 'sha384'
  • 'sha512'
  • 'keccak'

The digest encoding argument accepts the following values:

  • 'hex'
  • 'binary'

The hmac() method also supports 'base64' for the digest argument. This is for hmac() only, other implementations should use 'binary' with binaryToBase64().

Timestamps

All timestamps throughout all unified structures within this library are integer UTC timestamps in milliseconds!

In order to convert to milliseconds timestamps, CCXT implements the following methods:

const data = {
   'unixTimestampInSeconds': 1565242530,
   'unixTimestampInMilliseconds': 1565242530165,
   'unixTimestampAsDecimal': 1565242530.165,
   'stringInSeconds': '1565242530',
};

// convert to integer if the underlying value is already in milliseconds
const timestamp = this.safeInteger (data, 'unixTimestampInMilliseconds'); // === 1565242530165

// convert to integer and multiply by a thousand if the value has milliseconds after dot
const timestamp = this.safeTimestamp (data, 'unixTimestampAsDecimal'); // === 1565242530165

// convert to integer and multiply by a thousand if the value is a UNIX timestamp in seconds
const timestamp = this.safeTimestamp (data, 'unixTimestampInSeconds'); // === 1565242530000

// convert to integer and multiply by a thousand if the value is in seconds
const timestamp = this.safeTimestamp (data, 'stringInSeconds'); // === 1565242530000

Working With Array Lengths

In JavaScript the common syntax to get a length of a string or an array is to reference the .length property like shown here:

someArray.length

// or

someString.length

And it works for both strings and arrays. In Python this is done in a similar way:

len(some_array)

# or

len(some_string)

So the length is accessible in the same way for both strings and arrays and both work fine.

However, with PHP this is different, so the syntax for string lengths and array lengths is different:

count(some_array);

// or

strlen(some_string); // or mb_strlen

Because the transpiler works line-by-line and does no code introspection, it cannot tell arrays from strings and cannot properly transpile .length to PHP without additional hinting. It will always transpile JS .length to PHP strlen and will prefer string lengths over array lengths. In order to indicate an array length properly we have to do the following:

const arrayLength = someArray.length;
// the above line ends with .length;
// that ending is a hint for the transpiler that will recognize someArray
// as an array variable in this place, rather than a string type variable
// now we can use arrayLength for the arithmetic

That .length; line ending does the trick. The only case when the array .length is preferred over the string .length is the for loop. In the header of the for loop, the .length always refers to array length (not string length).

Adding Numbers And Concatenating Strings

In JS the arithmetic addition + operator handles both strings and numbers. So, it can concatenate strings with + and can sum up numbers with + as well. The same is true with Python. With PHP this is different, so it has different operators for string concatenation (the "dot" operator .) and for arithmetic addition (the "plus" operator +). Once again, because the transpiler does no code introspection it cannot tell if you're adding up numbers or strings in JS. This works fine until you want to transpile this to other languages, be it PHP or whatever other language it is.

There's this aspect of representation of numbers throughout the lib. The existing approach documented int the Manual says that the library will accept and will return "floats everywhere" for amounts, prices, costs, etc. Using floats is the easiest way of unboarding new users. This has known quirks, it's impossible to represent exact numbers with floats (https://0.30000000000000004.com/)

To address that, we are switching to string-based representations everywhere. So, we are now in the process of moving towards strings in a non-breaking way.

The new approach is:

We are adding an internal layer for string-based representations and string-based maths in the response parsers. That internal layer is built on top of the base Precise class, that is used to do all string-based maths. That class requires strings to operate on them and it will return strings as well. All existing old parsers must be rewritten to use Precise string-based representations, on first-encounter. All new code of all new parsers must be initially written with Precise string-based representations.

What exactly that means:

Compare this pseudocode showing how it was done before (an example of some arbitrary parsing code for the purpose of explaining it):

const amount = this.safeFloat (order, 'amount');
const remaining = this.safeFloat (order, 'remaining');
if (remaining > 0) {
    status = 'open';
} else {
    status = 'closed';
}
// ...
return {
    // ...
    'amount': amount,
    'remaining': remaining,
    'status': status,
    // ...
};

This is how we should do it from now on:

const amount = this.safeNumber (order, 'amount'); // internal string-layer
const remaining = this.safeString (order, 'remaining'); // internal string-layer
if (Precise.stringGt (remaining, '0')) { // internal string-layer
    status = 'open';
} else {
    status = 'closed';
}
// ...
return {
    // ...
    'amount': amount, // external layer, goes to the user
    'remaining': this.parseNumber (remaining), // external layer, goes to the user
    'status': status,
    // ...
};

In all new code of all parsers we should use string-based numbers throughout the body of the parser. Also we should add parseNumber as the last step of handling numeric values upon returning the result to the caller. The above two snippets are just examples, showing the usage of Precise with safeString and parseNumber. The actual parsers of orders also involve safeOrder-methods: https://github.com/ccxt/ccxt/pulls?q=safeOrder2.

The user will ultimately have an option to choose which implementation of parseNumber he wants: the one returning floats or the one returning strings. This way everyone will remain happy and the library will work both ways in a non-breaking fashion.

The rule of thumb is: + is for string concatenation only (!) and ALL arithmetic operations must use Precise.

Formatting Decimal Numbers To Precision

This section covers the request-assembly part. The .toFixed () method has known rounding bugs in JavaScript, and so do the other rounding methods in the other languages as well. In order to work with number formatting consistently use the decimalToPrecision method as described in the Manual.

Escaped Control Characters

When using strings containing control characters like "\n", "\t", always enclose them in double quotes ("), not single quotes (')! Single-quoted strings are not parsed for control characters and are treated as is in many languages apart from JavaScript. Therefore for tabs and newlines to work in PHP, we need to surround them with double quotes (especially in the sign() implementation).

Bad:

const a = 'GET' + method.toLowerCase () + '\n' + path;
const b = 'api\nfoobar.com\n';

Good:

const a = 'GET' + method.toLowerCase () + "\n" + path; // eslint-disable-line quotes
// eslint-disable-next-line quotes
const b = "api\nfoobar.com\n";

↑ The eslint-* comments are mandatory!

Using Ternary Conditionals

Do not use heavy ternary (?:) conditionals, always use brackets in ternary operators!

Despite that there is operator precedence in the programming languages themselves, the transpiler is regex-based and it does no code introspection, therefore it treats everything as plaintext.

The brackets are needed to hint the transpiler which part of the conditional is which. In the absence of brackets it's hard to understand the line and the intent of the developer.

Here are some examples of a badly-designed code that will break the transpiler:

// this is an example of bad codestyle that will likely break the transpiler
const foo = {
   'bar': 'a' + qux === 'baz' ? this.a () : this.b () + 'b',
};
// this confuses the transpiler and a human developer as well
const foo = 'bar' + baz + qux ? 'a' : '' + this.c ();

Adding surrounding brackets to corresponding parts would be a more or less correct way to resolve it.

const foo = {
   'bar': (qux === 'baz') ? this.a () : this.b (), // much better now
};

The universally-working way to solve it is to just break the complex line into a few simpler lines, even at a cost of adding extra lines and conditionals:

// before:
// const foo = {
//    'bar': 'a' + qux === 'baz' ? this.a () : this.b () + 'b',
// };
// ----------------------------------------------------------------------------
// after:
const bar = (qux === 'baz') ? this.a () : this.b ();
const foo = {
   'bar': 'a' + bar + 'b',
};

Or even:

// before:
// const foo = 'bar' + baz + qux ? 'a' : '' + this.c ();
// ----------------------------------------------------------------------------
// after:
let foo = 'bar' + baz;
if (qux) {
   foo += 'a';
};
foo += this.c ();

New Exchange Integrations

REMEMBER: The key reason why this library is used at all is Unification. When developing a new exchange file the goal is not to implement it somehow, but to implement it in a very pedantic, precise and exact way, just as the other exchanges are implemented. For that we have to copy bits of logic from other exchanges and make sure that the new exchange conforms to the Manual in the following aspects:

  • market ids, trading pair symbols, currency ids, token codes, symbolic unification and commonCurrencies must be standardized in all parsing methods (fetchMarkets, fetchCurrencies, parseTrade, parseOrder, ...)
  • all unified API method names and arguments are standard – can't add or change them freely
  • all parser input must be safe-sanitized as described above
  • for bulk operations the base methods should be used (parseTrades, parseOrders, note the s plural ending)
  • use as much of base functionality as you can, do not reinvent the wheel, nor the bicycle, nor the bicycle wheel
  • respect default argument values in fetch-methods, check if since and limit are undefined and do not send them to the exchange, we intentionally use the exchanges' defaults in such cases
  • when implementing a unified method that has some arguments – we can't ignore or miss any of those arguments
  • all structures returned from the unified methods must conform to their specifications from the Manual
  • all API endpoints have to be listed out with proper support for params substituted in the URLs

Please, see the following document for new integrations: https://github.com/ccxt/ccxt/wiki/Requirements

A quick merge of a Pull Request for a new exchange integration depends on consistency and compliance with the above unified rules and standards. Breaking one of those is the key reason for not merging a Pull Request.

If you want to add (support for) another exchange, or implement a new method for a particular exchange, then the best way to make it a consistent improvement is to learn from example. Take a look at how same things are implemented in other exchanges (we recommend certified exchanges) and try to copy the code flow and style.

The basic JSON-skeleton for a new exchange integration is as follows:

{
   'id': 'example',
   'name': 'Example Exchange',
   'country': [ 'US', 'EU', 'CN', 'RU' ],
   'rateLimit': 1000,
   'version': '1',
   'comment': 'This comment is optional',
   'urls': {
      'logo': 'https://example.com/image.jpg',
      'api': 'https://api.example.com/api',
      'www': 'https://www.example.com',
      'doc': [
         'https://www.example.com/docs/api',
         'https://www.example.com/docs/howto',
         'https://github.com/example/docs',
      ],
   },
   'api': {
      'public': {
         'get': [
            'endpoint/example',
            'orderbook/{pair}/full',
            '{pair}/ticker',
         ],
      },
      'private': {
         'post': [
            'balance',
         ],
      },
   },
}

Implicit API Methods

In the code for each exchange, you'll notice that the functions that make API requests aren't explicitly defined. This is because the api definition in the exchange description JSON is used to create magic functions (aka partial functions or closures) inside the exchange subclass. That implicit injection is done by the defineRestApi/define_rest_api base exchange method.

Each partial function takes a dictionary of params and returns the API response. In the example JSON above, the 'endpoint/example' results in the injection of a this.publicGetEndpointExample function. Similarly, the 'orderbook/{pair}/full' results in a this.publicGetOrderbookPairFull function, that takes a pair parameter (again, passed in the params argument).

Upon instantiation the base exchange class takes each URL from its list of endpoints, splits it into words, and then makes up a callable function name from those words by using a partial construct. That process is the same in JS, Python and PHP as well. It is also described here:

UNDER CONSTRUCTION

Docstrings

  • when a method takes another parameter as a property on params (ex. params['something']) add that parameter to the docstring, as params.something
    • if that parameter is required, the type is {str}, {int}, {etc}, if it's optional the type is {str|undefined}, {int|undefined}, {etc|undefined}
  • when a parameter's default value is undefined, but the method contains something like if (symbol === undefined) { throw new ArgumentsRequired('...')}, then set the type of that parameter as {str}, {int}, {etc}. If an error is not thrown, then the type is {str|undefined}, {int|undefined}, {etc|undefined}
  • if a method doesn't use one of the unified parameters, set the description of that parameter to not used by exchange_name.method_name () (replace exchange_name and method_name with the real exchange and method names)
  • if the method has any other special case uses, put these in the description of the docstring, these cases can be included in the class docstring as well

Continuous Integration

Builds are automated with Travis CI. The build steps for Travis CI are described in the .travis.yml file.

Windows builds are automated with Appveyor. The build steps for Appveyor are in the appveyor.yml file.

Incoming pull requests are automatically validated by the CI service. You can watch the build process online here: app.travis-ci.com/github/ccxt/ccxt/builds.

How To Build & Run Tests On Your Local Machine

Adding Exchange Credentials

CCXT has tests for both the public API and the private authenticated API. By default, CCXT's built-in tests will only test the public APIs, because the code repository does not include the API keys that are required for the private API tests. Also, the included private tests will not alter the balance of the account in any way, all tests are non-intrusive. In order to enable private API testing, one must configure the API keys. That can be done either in keys.local.json or with the env variables.

Configuring API keys and options in keys.local.json

Exchange API keys can be added to the keys.local.json in the root folder inside the repository. If it does not exist on your side – create it first. That file is in .gitignore and in .npmignore. You can add exchange credentials and various options for different exchanges to the keys.local.json file.

An example of keys.local.json file:

{
    "ftx": {
        "apiKey": "XXX",
        "secret": "YYY"
    },
    "binance": {
        "apiKey": "XXX",
        "secret": "YYY",
        "options": {
            "some-option": "some value"
        }
    },
    // ...
}
Configuring API keys as environment variables

You can also define API keys as env variables:

Consult the docs for your OS and shell on how to set an environment variable. Most of the time a set command, or a export command will work. The env command might help check the already-set environment variables.

Examples of env variables: BINANCE_APIKEY, BINANCE_SECRET, KRAKEN_APIKEY, KRAKEN_SECRET, etc.

Building

Before building for the first time, install Node dependencies (skip this step if you're running our Docker image):

npm install

The command below will build everything and generate PHP/Python versions from source JS files:

npm run build

Testing

The following command will test the built generated files (for all exchanges, symbols and languages):

node run-tests

You can restrict tests to a specific language, a particular exchange or symbol:

node run-tests [--js] [--python] [--python-async] [--php] [--php-async] [exchange] [symbol]

The node run-tests exchangename will try 5 tests: js, python, python-async, php, php-async. You can control that like so:

node run-tests exchange --js
node run-tests exchange --js --python-async
node run-tests exchange --js --php
node run-tests exchange --python --python-async
...

However, if that fails, you might have to bury one level lower and run language-specific tests to see what exactly is wrong:

node js/test/test exchange --verbose
python3 python/ccxt/test/test_sync.py exchange --verbose
python3 python/ccxt/test/test_async.py exchange --verbose
php -f php/test/test_sync.php exchange --verbose
php -f php/test/test_async.php exchange --verbose

The test_sync is just a synchronous version of test_async, so in most cases just running test_async.py and test_async.php is enough:

node js/test/test exchange --verbose
python3 python/ccxt/test/test_async.py exchange --verbose
php -f php/test/test_async.php exchange --verbose

When all of the language-specific tests work, then node run-tests will also succeed. In order to run those tests you want to make sure that the build has completed successfully.

For example, the first of the following lines will only test the source JS version of the library (ccxt.js). It does not require an npm run build before running it (can be useful if you need to verify quickly whether your changes break the code or not):

node run-tests --js                  # test master ccxt.js, all exchanges

# other examples require the 'npm run build' to run

node run-tests --python              # test Python sync version, all exchanges
node run-tests --php bitfinex        # test Bitfinex with PHP
node run-tests --python-async kraken # test Kraken with Python async test, requires 'npm run build'

Writing Tests

Follow this steps to add a test:

Committing Changes To The Repository

The build process generates many changes in the transpiled exchange files, e.g. for Python and PHP. You should NOT commit them to GitHub, commit only the base (JS) file changes please.

Financial Contributions

We also welcome financial contributions in full transparency on our open collective.

Credits

Contributors

Thank you to all the people who have already contributed to ccxt!

Backers

Thank you to all our backers! [Become a backer]

Supporters

Support this project by becoming a supporter. Your avatar will show up here with a link to your website.

[Become a supporter]

Sponsors

Thank you to all our sponsors! (please ask your company to also support this open source project by becoming a sponsor)

[Become a sponsor]