Skip to content

Latest commit

 

History

History
484 lines (432 loc) · 20 KB

CHANGELOG.md

File metadata and controls

484 lines (432 loc) · 20 KB
Raw

Changelog

This file does not aim to be comprehensive (you have git history for that), rather it lists changes that might impact your own code as a consumer of this library.

3.0.0-beta.10

New additions

  • takeWhile - returns a new stream that ends when the function passed as a parameter stops returning true. #677.
  • Async Iterator/Iterable support - Taks an object that is a async iterator or iterable as defined by the Async Iteration specs. #682.

3.0.0-beta.9

This release contains all changes from 2.13.3.

3.0.0-beta.8

This release contains all changes from 2.13.1.

3.0.0-beta.7

New additions

Bugfix

  • of - Streams created with of didn't contain extentions added by use. #662 Fixes #659.

3.0.0-beta.6

This release contains all changes from 2.12.0.

Breaking changes

  • toNodeStream - In 2.x, if you call toNodeStream on a Highland stream that emits objects without setting {objectMode: true}, the resulting Readable may simply ignore any objects that are emitted (depending on the implementation of StringDecoder). In 3.x, the Readable will always emit an error.

    Also, in 2.x, calling toNodeStream will immediately consume the source. In 3.x, the source stream will only be consumed when the Readable is consumed.

Other

  • stopOnError - update stopOnError documentation to include push #650.

3.0.0-beta.5

Breaking changes

  • stream.source - The (undocumented) source property on a stream no longer exists.
  • consume - It is no longer possible to consume the same stream multiple times, even after the original consume has pushed nil. Attempting to do so will result in an error. This affects all transforms that are implemented via consume.
    • Old Behavior: Code like this used to work 
      const stream = _([1, 2, 3]);
stream.take(1)
  .toArray(array => {
    console.log(array); // => [1]
    stream.take(1).toArray(array2 => {
      console.log(array2); // => [2]
      stream.take(1).toArray(array3 => {
        console.log(array3); // => [3]
      });
    });
  });
    • New Behavior: 
      const stream = _([1, 2, 3]);
stream.take(1)
  .toArray(array => {
    console.log(array); // => [1]
    stream.take(1); // The call to take(1) will fail.
  });
  • drop #559 Fixes #594
    • Old Behavior: if n isn't a number or is negative, it defaults to 0.
    • New Behavior: if n is not a number or if it is negative, an error is thrown.
  • fork - It is no longer possible to call fork after a call to consume on the same stream. Attempting to do so will result in an error. This also applies to all transforms. Note that each transform returns a new stream, so it is still possible to fork a stream that has been transformed.
    • Code like this used to work: 
      const stream = _([1, 2, 3]).map(x => x + 1);
const fork1 = stream.map(x => x * 2);        // stream is consumed.
const fork2 = stream.fork().map(x => x * 3); // then forked.
// fork1 and fork2 share backpressure.
      In 3.0.0, this code should be written as 
      const stream = _([1, 2, 3]).map(x => x + 1);
const fork1 = stream.fork().map(x => x * 2); // stream must be explicitly forked
const fork2 = stream.fork().map(x => x * 3);
  • map - Passing a non-function value to map now throws an error.
    • Old Behavior - Passing a non-function value to map caused it to map all stream values to the passed in value.
  • reduce - The order of the arguments to reduce has been swapped.
    • Old Behavior: stream.reduce(memo, reducer)
    • New Behavior: stream.reduce(reducer, memo)
  • scan - The order of the arguments to scan has been swapped.
    • Old Behavior: stream.scan(memo, reducer)
    • New Behavior: stream.scan(reducer, memo)
  • slice #559 Fixes #594
    • Old Behavior: if start isn't a number, it defaults to 0, and if end isn't a number, it defaults to Infinity. If start or end are negative, they are treated as if they are 0 instead.
    • New Behavior: if start or end are not numbers or if they are negative, an error is thrown. start and end are optional and default to 0 and Infinity, respectively.
  • take #559 Fixes #594
    • Old Behavior: if n isn't a number, it defaults to Infinity. If n is negative, it is treated as if it is 0.
    • New Behavior: if n is not a number or if it is negative, an error is thrown.
  • zipAll - zipAll has been renamed zipEach.
  • zipAll0 - zipAll0 has been renamed zipAll.

New additions

  • wrapAsync: Wraps a function that returns a promise, transforming it to a function which accepts the same arguments and returns a Highland Stream instead. #548. Fixes #517.

2.13.3

Bugfix

  • Fixes a bug in pipe where errors would be emitted twice.

2.13.2

Bugfix

  • Fixes a regression in the through transform. If it is called with an un-paused through stream and the source Highland stream has buffered data (or generates data synchronously), then that data may be lost. #671.
  • Fix pipe so that it does not emit data synchronously. This allows for chained pipes to work properly even if a stream in the middle of the pipe is unpaused. #671.

2.13.1

Bugfix

  • Fixes a potential deadlock when wrapping node Readables in node 10+. #670.

2.13.0

New additions

  • isNil: Returns true if the argument is the end of stream marker. This can be useful as a user-defined type guard in Typescript. #645

2.12.0

New additions

  • toNodeStream: Returns a native node Readable #644

2.11.1

Bugfix

  • Remove usages of Function.prototype.bind. We support IE8. #632.
  • Add a section about supported JS engines to the documentation.

2.11.0

New additions

  • toPromise: Converts a one-element or zero-element stream to a Promise given a Promise/A+ compliant promise constructor. #628. Fixes #627.

2.10.5

Bugfix

  • Streams constructed from an EventEmitter or jQuery element will now remove itself when the stream is destroyed. Fixes #500. #609.

2.10.4

Bugfix

  • Same as 2.10.3 but with a more conservative fix to minimize the risk of regressions.

2.10.3

Bugfix

  • In certain cases, consuming a stream that has been resumed may cause a stream generator/consume handler to be called twice without next() ever being called. This is mostly relevant for .each(...).done(...) use cases. Noticed in #570 (comment). #608.

2.10.2

Bugfix

  • Uncaught errors from promise-back streams weren't being correctly logged in certain circumstances when using a Promise implementation that does not log unhandled promise exceptions. All uncaught highland errors should now be correctly logged. #591. Fixes #589.
  • Users using bluebird as their Promise implementation may have seen an error that says "a promise was created in a handler at ... but was not returned from it". This is a false positive, and Highland's use of promises have been updated to suppress this warning. #588.

2.10.1

Bugfix

  • Asynchronously pushing a nil in consume when then input value wasn't a nil itself now no longer causes the stream to deadlock. #564. Fixes #563. Related to #558.
  • Much improved documentation. Examples are now more standalone, and more guidance was added for certain common pitfalls.

2.10.0

New additions

  • of: Creates a stream that sends a single value then ends. #520.
  • fromError: Creates a stream that sends a single error then ends. #520.
  • When constructing a Highland stream from a Node Readable, the onFinish handler may now turn off the default automatic end on errors behavior by returning an object with the property continueOnError set to true. #534. Fixes #532.

2.9.0

New additions

  • It is now possible to pass an custom onFinish handler when constructing a Highland Stream from a Node Readable Stream. This allows for special detection of stream completion when necessary. #505. See #490 for a discussion on why this is necessary.

2.8.1

Bugfix

  • The Readable stream wrapper changes from 2.8.0 assumed that close would never be emitted before end for any stream. This is not the case for Sockets, which will close when the client disconnects but will end only when it has piped all of its data. For a slow consumer, end may happen after close, causing the Highland stream to drop all data after close is emitted.

    This release fixes the regression at the cost of restoring the old behavior of never ending the Stream when only close is emitted. This does not affect the case where error events are emitted without end. That still works fine. To manually end a stream when it emits close, listen to the event and call stream.end(). Fixes #490.

2.8.0

Bugfix

  • A Highland Stream that wraps Readable now properly handles the case where the Readable emits the close event but not the end event (this can happen with an fs read stream when it encounters an error). It will also end the wrapper stream when it encounters an error (this happens when reading from a non-existent file). Before, such streams would simply never end. #479. Fixes #478.

New additions

  • toCallback: method for returning the result of a stream to a nodejs-style callback function. #493. Fixes #484.

Improvements

  • A Highland Stream that wraps a bluebird promise can now handle bluebird cancellation. When the promise is cancelled the wrapper stream is empty. #487. Fixes #486.

2.7.4

Bugfix

  • mergeWithLimit no longer causes an // Unhandled 'error' event error when one of its sources emits an error. #476. Fixes #475.

2.7.3

Bugfix

  • pipe now properly unbinds its drain handler from the destination when it is done. Previously, there would have been a memory leak if the destination is long-lived (e.g., as with process.stdout). #466.

2.7.2

Bugfix

  • Minor fixes to the documentation.

New additions

  • The library's browserify bundle is now published to NPM alongside the regular code. #310. Fixes #309.

2.7.1

Bugfix

  • pipe now emits the pipe event to the destination stream. #450. Fixes #449.

New additions

  • pipe now takes a second, optional options argument that allows users to decide whether or not to end the destination stream when the source ends. #450.

2.7.0

Broken release. Use 2.7.1 instead.

2.6.0

Bugfix

  • parallel no longer drops elements on the floor in a number of cases. #302, #331. Fixes #234, #328.
  • Calling next before push within a generator stream no longer causes the stream to resume and throw away data when used with pull. #326. Fixes #325.
  • Parallel no longer drops data if paused. #331. Fixes #328.
  • Various grammar fixes and documentation updates. #341, #354, #381, #397, #407
  • isStream now always returns a boolean. Before, it would return undefined if the argument was an object but not a Highland stream. #343.
  • Streams now unpipe from Readable on destroy. #361.
  • _send now keeps a reference to the correct consumer/observer array. #367. Fixes #366.
  • Streams constructed with pipeline now correctly exert backpressure. #372, #377. Also fixes an possible issue with not consuming errors from promises. #391.
  • It is no longer possible to re-enter the consume callback. #393.

New additions

  • mergeWithLimit: Like merge, but with an argument to specify the maximum number of parallel stream that can be consumed at once. #375.
  • minified build: There is now a minified version of the browser build under dist/highland.min.js. #392.
  • wrapCallback: The function now takes a second argument (mappingHint) that describes how arguments passed to the callback are handled. It behaves like the mappingHint parameter of the stream constructor. #247. Fixes #246, #334.
  • Node 4 and 5: Added support for node 4 and 5. #383.

Improvements

  • The runtime of pick per object is now O(n), where n is the number of properties to be picked. It was previously O(mn), where m is the number of pickable properties on the object. #286.
  • Both pick and pickBy can now select non-enumerable keys. #286.
  • parallel now throws descriptive errors if it encounters a value that is not a stream. #318.
  • The standalone Highland file is now built using Browserify 12.0.1.
  • Updates a number of devDependencies. If you develop on Highland, make sure to update the dependencies. #384, #385, #387, #390, #400, #403, #415.
  • uniq now uses a Set to compute uniqueness whenever available, resulting in a significant performance boost for large streams. The definition of equality is still ===, not the SameValueZero algorithm used by Set. #395
  • parallel now throws if provided an argument that is not a number. #421.

Other

  • Dropped support for Node 0.11.
  • Dropped support for iojs.
  • Deprecation warnings for API changes upcoming in version 3.0.0 have been added. #417

2.5.1

Bugfix

  • Move stream check in constructor to beginning of object branch. #303

2.5.0

New additions

  • drop: Ignores the first n values of a stream and then emits the rest. #75 #244
  • done: Calls the supplied function once the stream has ended. #161
  • sort: Collects all values together then emits each value individually but in sorted order. #169 #245
  • streamifyAll: Takes an object or a constructor function and returns that object or constructor with streamified versions of its function properties. #226
  • Iterator Support: ECMA2015 (aka ES6) style iterators can now be passed to the Highland constructor function. #235
  • slice: Creates a new stream with the values from the source in the range of specified in thestart and end parameters. #250
  • batchWithTimeOrCount: Takes one Stream and batches incoming data within a maximum time frame into arrays of a maximum length. #284

Improvements

  • each now returns an empty stream rather than nothing. #161.
  • Ensure through propagates Node stream errors. #240
  • Preserve this context of wrapped function when using wrapCallback. #248
  • Update tranduce to use latest version of transformer protocol. #261

2.0.0

  • The source.merge() algorithm now evaluates the entire source stream before reading from all of the resulting streams in parallel (previously it would start reading as soon as the source emitted the next stream)
  • The merge() function now attempts to balance inputs more fairly. For example, if stream A has 100 values buffered and stream B gets a new value after 100ms, if we read at 200ms we'll get a value from each stream. Previously it would exhaust the stream A buffer before reading from stream B.