We're working hard to make berycore the most powerful JavaScript library for working with berycoin. Our goal is to have berycore be a library that can be used by anyone interested in berycoin, and to level expertise differences with great design and documentation.
If there are any questions, etc., please feel to ask in one of the community channels:
- https://labs.bitpay.com/c/bitcore (Support Forum)
- https://gitter.im/bitpay/bitcore (Development Chat)
Ideally, please make sure to run:
gulp testpasses all the tests (We run tests against Node.js v0.10, v0.12, io.js, and modern browsers)gulp coveragecovers 100% of the branches of your code (Seecoverage/lcov-report/index.htmlfor details)gulp lintdoesn't complain about your changes
These are some global design goals in berycore that any change must adhere.
We take our time with picking names. Code is going to be written once, and read hundreds of times.
We were inspired to name this rule first due to Uncle Bob's great work Clean Code, which has a whole chapter on this subject.
In particular, you may notice that some names in this library are quite long for the average JavaScript user. That's because we prefer a long but comprehensible name than an abbreviation that might confuse new users.
Write a test for all your code. We encourage Test Driven Development so we know when our code is right. We have increased test coverage from 80% to around 95% and are targeting 100% as we move towards our 1.0 release.
Be conservative in what you send, be liberal in what you accept.
Interfaces should accept as many types of arguments as possible, so there's no mental tax on using them: we want to avoid questions such as "should I use a string here or a buffer?", "what happens if I'm not sure if the type of this variable is an Address instance or a string with it encoded in base-58?" or "what kind of object will I receive after calling this function?".
Accept a wide variety of use cases and arguments, always return an internal form of an object. For example, the class PublicKey can accept strings or buffers with a DER encoded public key (either compressed or uncompressed), another PublicKey, a PrivateKey, or a Point, an instance of the elliptic.js library with the point in berycoin's elliptic curve that represents the public key.
Consistency on the way classes are used is paramount to allow an easier understanding of the library.
The design guidelines have quite a high abstraction level. These style guidelines are more concrete and easier to apply, and also more opinionated. The design guidelines mentioned above are the way we think about general software development and we believe they should be present in any software project.
Follow this Node.js Style Guide: https://github.com/felixge/node-style-guide#nodejs-style-guide
Avoid constants in the code as much as possible. Magic strings are also magic numbers.
If a class has a publicKey member, for instance, that should be a PublicKey instance.
Avoid representation errors by always dealing with satoshis. For conversion for frontends, use the Unit class.
A special case for G2 all network references must be Network instances (see lib/network.js), but when returned to the user, its .name property should be used.
Write a .inspect() method so an instance can be easily debugged in the console.
Name them in UpperCamelCase, as they are namespaces.
DO:
var BufferUtil = require('./util/buffer');DON'T:
var bufferUtil = require('./util/buffer');When possible, berycore objects should have standard methods on an instance prototype:
toObject/toJSON- A plain JavaScript object thatJSON.stringifycan calltoString- A string representation of the instancetoBuffer- A hex Buffer
These should have a matching static method that can be used for instantiation:
fromObject- Should be able to instantiate with the output fromtoObject/toJSONfromString- Should be able to instantiate with output fromtoStringfromBuffer- Should likewise be able to instantiate from output fromtoBuffer
JSON.stringify and JSON.parse are expected to be handled outside of the scope of Berycore methods. For example, calling JSON.stringify on a Berycore object will behave as expected and call transaction.toJSON() and then stringify it:
var transactionString = JSON.stringify(transaction);Likewise to instantiate a transaction from that string:
var data = JSON.parse(transactionString);
var tx = new Transaction(data);We've designed a structure for Errors to follow and are slowly migrating to it.
Usage:
- Errors are generated in the file
lib/errors/index.jsby invokinggulp errors. - The specification for errors is written in the
lib/errors/spec.jsfile. - Whenever a new class is created, add a generic error for that class in
lib/errors/spec.js. - Specific errors for that class should subclass that error. Take a look at the structure in
lib/errors/spec.js, it should be clear how subclasses are generated from that file.
In order to deal with JavaScript's weak typing and confusing errors, we ask our code to fail as soon as possible when an unexpected input was provided.
There's a module called util/preconditions, loosely based on preconditions.js, based on guava, that we use for state and argument checking. It should be trivial to use. We recommend using it on all methods, in order to improve robustness and consistency.
$.checkState(something === anotherthing, 'Expected something to be anotherthing');
$.checkArgument(something < 100, 'something', 'must be less than 100');
$.checkArgumentType(something, PrivateKey, 'something'); // The third argument is a helper to mention the name of the argument
$.checkArgumentType(something, PrivateKey); // but it's optional (will show up as "(unknown argument)")Most classes have static methods named fromBuffer, fromString, fromJSON. Whenever one of those methods is provided, the constructor for that class should also be able to detect the type of the arguments and call the appropriate method.
For classes that have a mutable state, most of the methods that can be chained SHOULD be chained, allowing for interfaces that read well, like:
var transaction = new Transaction()
.from(utxo)
.to(address, amount)
.change(address)
.sign(privkey);Constructors, when provided an instance of the same class, should:
- Return the same object, if the instances of this class are immutable
- Return a deep copy of the object, if the instances are mutable
Examples:
function MyMutableClass(arg) {
if (arg instanceof MyMutableClass) {
return MyMutableClass._deepCopy(arg);
}
// ...
}
function ImmutableClass(arg) {
if (arg instanceof ImmutableClass) {
return arg;
}
// ...
}Constructors should not require to be called with new. This rule is not heavily enforced, but is a "nice to have".
function NoNewRequired(args) {
if (!(this instanceof NoNewRequired)) {
return new NoNewRequired(args);
}
// ...
}Style guidelines are not relaxed for tests. Tests are a good way to show how to use the library, and maintaining them is extremely necessary.
Don't write long tests, write helper functions to make them be as short and concise as possible (they should take just a few lines each), and use good variable names.
Inputs for tests should not be generated randomly. Also, the type and structure of outputs should be checked.
This helps to make tests more useful as examples, and more independent of where they are placed. This also helps prevent forgetting to include all submodules in the berycore object.
DO:
var berycore = require('../');
var PublicKey = berycore.PublicKey;DON'T:
var PublicKey = require('../lib/publickey');If possible, data for tests should be included in a JSON file in the test/data directory. This improves interoperability with other libraries and keeps tests cleaner.
All modules should include a developer guide and API reference. The API reference documentation is generated using JSDOC. Each function that exposes a public API should include a description, @return and @param, as appropriate. The general documentation guide for the module should be located in the docs/guide directory and is written in GitHub Flavored Markdown.
Please proofread documentation to avoid unintentional spelling and grammatical mistakes before submitting a pull request.
Our workflow is based on GitHub's pull requests. We use feature branches, prepended with: test, feature, fix, refactor, or remove according to the change the branch introduces. Some examples for such branches are:
git checkout -b test/some-module
git checkout -b feature/some-new-stuff
git checkout -b fix/some-bug
git checkout -b remove/some-fileWe expect pull requests to be rebased to the master branch before merging:
git remote add berycoin-project git@github.com:berycoin-project/berycore.git
git pull --rebase berycoin-project masterNote that we require rebasing your branch instead of merging it, for commit readability reasons.
After that, you can push the changes to your fork, by doing:
git push origin your_branch_name
git push origin feature/some-new-stuff
git push origin fix/some-bugFinally go to github.com/berycoin-project/berycore in your web browser and issue a new pull request.
Main contributors will review your code and possibly ask for changes before your code is pulled in to the main repository. We'll check that all tests pass, review the coding style, and check for general code correctness. If everything is OK, we'll merge your pull request and your code will be part of berycore.
If you have any questions feel free to post them to github.com/berycoin-project/berycore/issues.
Thanks for your time and code!