stream: simpler stream constructon

[sonewman]

This allows stream implementers the advantage of being able to pass the necessary stream specific methods to the streams constructor as part of its options upon construction.

Referenced to discussion in issue nodejs/readable-stream#102 of iojs/readable-stream

Please review.

cc: @chrisdickinson @iojs/streams

[domenic]

Technically I wouldn't call this a revealing constructor pattern, since no functionality is "revealed"; you instead just use stream.push and stream.emit("error", ...) and similar.

[sonewman]

I get your point, I will amend the description as well.

[Qard]

+1 I like it. It makes the public-facing part of streams somewhat more approachable. They're currently a bit of a nebulous thing to newbies, hence the existence of through/through2.

[sonewman]

@domenic i have updated.

[domenic]

s/the need of//

[domenic]

LGTM with minor nits. Someone more familiar with testing practices in io.js should sign off on those.

[sonewman]

@domenic I appreciate your critique. I've never been that great at writing copy 😃

[Qard]

The console.log('ok') bits don't need to be there, and I'd refactor the writable test to use one process.on('exit', ...) and no unnecessary one/two functions.

[mafintosh]

I would remove new from the examples since that is optional. Other than that LGTM from me

[domenic]

new is not optional for ES6 classes so it'd be prudent to leave them in to discourage bad habits.

[sonewman]

@Qard I was trying to follow the style of the other stream tests... But I will amend.

[chrisdickinson]

@domenic This is more for my curiosity than anything else: can you still get around that with the usual if (!(this instanceof Classname)) return new Classname(...args) in constructor(){}? If that's the case, I imagine we'll continue to take the "new is encouraged but not required" approach.

[chrisdickinson]

Should we reframe how we're documenting this? Would it be better to integrate the existing examples with the new style of construction, and call out "old-style subclassing" as the standalone topic?

[domenic]

@chrisdickinson nope, constructors defined with class syntax have a throwing [[Call]] internal method.

[sonewman]

Should we reframe how we're documenting this? Would it be better to integrate the existing examples with the new style of construction, and call out "old-style subclassing" as the standalone topic?

@chrisdickinson I thought about this very thing, as I wasn't sure where the documentation should belong. If it went before the inheritance parts then the examples were referring to documentation that was to come. By putting it after solved that problem, but one would have to read the older inheritance way to understand the use of this new simpler pattern. 😕

[chrisdickinson]

@sonewman Cool. We should definitely revisit the docs later, but that shouldn't be considered blocking for this.

I'll leave this open until tomorrow afternoon, at which point I'll merge it if no one has objected.

[sonewman]

@chrisdickinson awesome cheers dude.

[vkurchatkin]

👍 wanted this for a long time

[sonewman]

@chrisdickinson do you want me to squish this into one commit?

[Qard]

This test is failing for me:

=== release test-stream-writable-constructor-set-methods ===                   
Path: parallel/test-stream-writable-constructor-set-methods
assert.js:87
  throw new assert.AssertionError({
        ^
AssertionError: null == [Function: _writev]
    at process.<anonymous> (/home/sbelanger/Documents/io.js/test/parallel/test-stream-writable-constructor-set-methods.js:31:10)
    at process.emit (events.js:119:20)
Command: out/Release/iojs /home/sbelanger/Documents/io.js/test/parallel/test-stream-writable-constructor-set-methods.js
[00:43|% 100|+ 812|-   1]: Done                                                

[chrisdickinson]

Looks like a typo. Fixing in the merge.

[chrisdickinson]

Merged in 50daee7.

Note: This is a semver-minor level change. If we don't want the next release to be semver-minor, this should be backed out first.

[sonewman]

@chrisdickinson thanks for fixing the typo, I don't know how many times I ran those tests (but obviously not enough!)