redux-saga 1.0.0

v1.0.0

on Node.js NPM

Breaking changes:

• channel and actionChannel have default buffer of buffers.expanding()
• errors thrown during put execution are no longer caught and swallowed, you need to catch them manually
• signature of join & cancel, they both accept now a single task descriptor or an array of those (previously they have accepted variadic length of arguments)
• removed some deprecated APIs - takeEvery, takeLatest, throttle from the redux-saga entry point (they are and were importable from redux-saga/effects), takem, put.sync and executing array of effects, there is explicit API for this for already some time - all effect
• changed API of runSaga - it no longer accepts subscribe option, you should create a channel (preferably stdChannel), pass it as channel argument to the runSaga API and communicate with through it with take and put methods
• refactored shape of the effect objects to { [IO]: true, type, payload }
• removed asEffect, effect types are public, effect shapes are stable, so if for whatever reason you want to do some effect inspection you can just check them without us providing additional helpers for it
• removed logError, use only onError option (it's signature is onError(error, { sagaStack }))
• END will now finish the race effects
• task.done getter was changed to be task.toPromise method
• channels private getters (__takers__ and __closed__) got removed
• delay became an effect
• redux-saga/utils got removed, exports available exclusively there got moved to separate packages - @redux-saga/deferred, @redux-saga/delay-p, @redux-saga/is, @redux-saga/symbols & @redux-saga/testing-utils
• eventChannel does no longer accept matcher argument - it was a hacky way to support previous implementation of stdChannel
• {effects, utils} can't imported from 'redux-saga' anymore
• most runtime type checks got hidden behing development checks, inputs might not be validated in production (failed validation resulted in error being thrown anyway)
• detach helper returns a new effect instead of mutating the input one
• we have stopped interpreting effects returned from fork, this shouldn't affect any real-life code, affected patterns look like this fork(() => effectCreator()) and fork(takeEvery, 'type', fn)
• TS typings got revamped - they require TS@>3.1 now

New:

• babel-plugin-redux-saga - can be used to enhance stack traces of thrown errors
• multicastChannel - no buffering, notify all pending takers, multicastChannel#take(cb, matcher = matchers.wildard)
• support for yield take(multicastChannel, pattern)
• internal stdChannel got reworked to be a singleton object (it is wrapped multicastChannel's instance'), also it is an exported API to support new runSaga's signature - this should also result in being a small perf boost
• effectMiddlewares - useful especially for testing, you can intercept/hijack any effect and resolve it on your own - passing it very redux-style to the next middleware (last being redux-saga itself). How it might be used can be checked here.
• takeLeading effect - it takes "leading" action and ignores all incoming ones of the same type while the "leading" is still handled (useful for things debouncing)
• retry effect with the signature of retry(maxTries, delayLength, worker, ...args)
• debounce effect with the signature of debounce(delayLength, pattern, worker, ...args)
• new rootSagaStarted hook for saga monitor
• added dev warning about dispatching frozen actions, for scheduling purposes we are using Object.defineProperty on actions dispatched with put effect, so we have to be able to mutate the action object
• added dev warning about using async generators, they are not supported by redux-saga
• CommonJS entries are proxied now. This means that process.env.NODE_ENV is read only once and based on that value the appropriate bundle (development or production) is loaded conditionally with require. process.env is slow in node.js so this should improve performance a little bit.
• isRoot property on Task (you probably won't ever need to use it, it's internal)

Fixes:

• keeping single stdChannel in the internals allowed to fix 2 bugs with missed actions (see #707 and #1146)
• onError should get called now even if you throw non-Errors in your code.
• we suspend the scheduler before running a root saga, so it should behave the same as forked sagas in terms of scheduling
• small memory leak for closed actionChannels
• issue with joiner not ending in cancelled state when joining cancelled task