link May 22 Meeting Notes

(by Erik Arvidsson)

John Neumann (JN), Allen Wirfs-Brock (AWB), Eric Ferraiuolo (EF), Erik Arvidsson (EA), Luke Hoban (LH), Doug Crockford (DC), Yehuda Katz (YK), Brendan Eich (BE), Sam Tobin-Hochstadt (STH), Alex Russell (AR), Dave Herman (DH) (calling in), Bernd Mathiske (BM), Andreas Rossberg (ARB), Mark Miller (MM), Tom Van Cutsem (TVC), Istvan Sebestyen (IS), Jasvir Naga (JNA)

link 4.16 Spec update

http://wiki.ecmascript.org/doku.php?id=harmony:specification_drafts

YK: ToPositiveInteger is needed by JSIDL

AI(YK+AWB): Put an algorithm in the spec that DOM can use so that we get the same behavior in JS and DOM.

6 General implementation experiences

ARB: We started implementing generators but things are pretty smooth.

BE: Doing modules at the moment.

AWB: Bunch of bug fixes in the spec related to classes.

link 4.9 Template Strings (Template Literals)

MM: Suggests status quo.

AR: Objects

MM: Controversy related to tag-less templates. Alternatives include making tag-less templates an error, delayed evaluation (contextually provided)

AR: Econics: naked interpolation is too attractive. Should always have a tag to encourage users to think about which behavior is correct.

YK: I cannot support Alex's proposal.

STH: What would the name of this tag be?

AR: Something that is imported from a module.

STH: Concerned about short names and conflicts.

YK: People will just use 's' without thinking.

YK: People should use HTML templating engines.

DC: Alex's testimony about application developer feedback is relevant.

LH: it sounded like Google engineers were using a template system

EA: Correct.

MM: Does anyone prefer taking out TS if they don't get tag-less TS?

Everyone: Agrees that it is better to require tag than to remove TS from ES6.

AR: Strings are always used later in some context. Communicating the intent

AWB: String concat vs string interpolation have the same issue.

LH: Assumes that maybe only 20% of the uses of TS are susceptible to XSS

MM: Removing tag-less does not reduce XSS because people will just use

s`...`

TS helps people transition to a better world. Once they have have a TS it will be easy to add an html tag at the front as needed.

ST: It will be painful to import String raw and alias that to s.

MM: Maybe put tag-less in appendix? Withdrawn idea because no one likes it.

YK: You should not have use string based APIs.

AR: Willing to abstain but "Y'all are making a big mess"

BM: Half convinced by Alex.

LH: Different code bases will use different tags for normal string interpolation so moving between code bases will be hard to.

AR: That is a good thing. Forces people to think.

MM: Template strings in E.

STH: Lots of contexts where XSS is not an issue.

BM: More ways to XSS is a bad thing.

BE: if people have to import s then the economics change and people will stick to +

link Consensus/Resolution:

• AR and BM sustains.
• Continue with the status quo (tag-less TS is supported)

link JSON

DC: IETF wants to change JSON

MM: The 2 documents should have exactly the same text except for boilerplate.

IS: Should it be done in TC39?

DC: Most of the work will be on the mailing lists

AWB: Who will be the editor?

DC: Hopes they (IETF) will provide an editor.

JN: Should this be fast tracked to ISO?

DC: That makes sense.

JN: How long do you expect this to take?

DC: Has taken a long time to coordinate and get started. 5.1 specs the 2 functions that uses the JSON format.

link 4.10 Modules

STH: Progress since last meeting. Discuss "module naming", "naming standard modules". http://wiki.ecmascript.org/doku.php?id=harmony:modules Wiki is up to date with the current proposal. Spec is "wiki complete".

Jason Orendorff of Mozilla has worked on flushing out semantic issues. Moz is implementing parsing of modules.

STH: Syntax: Made a couple of changes.

A. To support anonymous exports

export default expr;
 
import $ from "jquery";  // imports default anonymous export

If there is no default then the above is an error

import {ajax} from "jquery";
 
import {ajax as A} from "jquery";

to reduce confusion and to make it clear that this is not destructuring.

module fs from "js/fs"

• fs is a module instance object

The following is not valid:

import {...} from fs;  // SyntaxError

Renaming on export:

let foo = 13;
export { foo as bar };
export { foo };

The following is not valid:

export foo;

STH: The only evaluation here is "13". The rest are just bindings that are shared with the outside/module importer.

MM: Bad idea to allow external modules to assign to imports.

DH: Imported bindings are read only to the importer.

AWB: This is new semantics to the language. Is there a list of these new semantics modules introduce?

AWB: Is there a way to get the default export from the instance module obejct.

STH: There will be a well known symbol name to get to it.

AWB: Does module instance objects inherit from Object.prototype.

DH: No. Because we do not want any pollution.

JNA: Is it an error to assign to an imported binding?

import {ajax} from "jquery";
ajax = 14;  // Error

AR: What is the reason for not extending Object.prototype or some other object?

YK: To prevent people from expecting toString to be there (???)

DH: fs.readFile We don't want to statically check this deeply inside an expression.

fs.toString

THS: The plan is to allow the above to be a static error in the future.

DH: To keep things clean.

AWB: Concerned about the dot operator

ARB: Don't want less checking if you do not use import.

DH: Do not want refactoring hazards.

ARB: This only affect the static semantics.

AWB: Can you use square bracket?

STH: Square bracket is dynamic.

AR: This is only a static check that is lost. At runtime there will still be errors.

LH: Concerned about default export. Now people will have to decide which approach to use.

STH: This is already the case in Node.js today.

LH: Today you might get any object, it might be callable with properties.

var fs = require("fs");  // module instance
var glob = require("glob");  // function with properties
var parse = require("parse");  // function
 
module fs from "fs";
import glob from "glob";
import {sync} from "glob";
import parse from "parse";

Lots of discussion...

import {sync} from "glob";

alt

import glob from "glob";
var {sync} = glob;
import {ajax} from "jquery";

LH: Prefers export = and lose static checking when people opt in to single anonymous export.

STH/YK: We already agreed that we want static checking.

LH: Even for new things being built, this is causing a confusion.

AWB: It is unclear when and what you want to export as the default export.

BM: Wants

import default $ from "jquery"

...to ensure that people have to be explicit about what they import.

DH: This is just syntax and we are wasting time "bikeshedding"

AWB: What is the best practice? Is there a single module containing Map, Set & WeakMap or...

YK: WeakMap should be its own import:

import WeakMap from "collections/WeakMap";

BE: We have to pay attention to what Node/AMD do today.

YK: AMD tries to make modules small to reduced byte size of the dependencies.

STH: And now to semantics https://github.com/jorendorff/js-loaders/blob/master/browser-loader.js

Major things that changed. Use options object more consistently. The wiki page is up to date. Need to decide whether the browser loader is in the appendix or if it is in some w3c spec. Want core language semantics to treat the names as strings, not the semantics of these strings. Bulk loading. One HTTP request to load multiple modules. Possible to implement. Create fecth hook. Stores module notations in a side table. In the xhr response, split the result and call the different fulfill hooks.

EF: Sounds like what we do today in YUI loaders. How would you write the HTML?

DH: Initial script tag with configuration. Second script tag as usual. Alt 2 is to have configuration and dynamic module load in the same script block.

<script>
ondemand
</script>
<script src="main.js" async></script>

alt 2

<script>
ondemand
System.require("main.js", function() { .... });
</script>

DH: script[async] today have to use an external src.

STH: Naming and declarations of modules.

ARB: Presenting slides...

AWB: The rate that internal vs external names changes is very different.

STH:

module "m" { ... }
module "n" {
  import x from "m";
  ...// this part is not executed.
}
import x from "m";

STH: Configuration step is mostly about other people's code.

<script>
module "m" { ... }
module "n" {
  import m from "m";
  function f() {
    Loader.eval("import m from 'n'");
  }
}
</script>

m is fixed at compile time

ARB: Not opposed to logical modules. Wants both lexical and logical

DH: Not opposed to lexical modules.

YK: Too late to work out lexical modules for ES6.

ARB: If we wait we will have redundancy.

YK: Want declarative form to be able to prefetch etc.

BE: I want lexical modules (in the future) but logical modules are easier to use.

ARB: Since I don't seem to be able to convince anyone I'm going to drop this

ARB: For the record. Major concern about the global registry becoming the new global object.

link Consensus/Resolution:

• Move along with Dave and Sam's proposal.
• Work on lexical modules for ES7

link May 21 Meeting Notes

(by Erik Arvidsson)

John Neumann (JN), Allen Wirfs-Brock (AWB), Eric Ferraiuolo (EF), Erik Arvidsson (EA), Luke Hoban (LH), Doug Crockford (DC), Yehuda Katz (YK), Brendan Eich (BE), Sam Tobin-Hochstadt (STH), Alex Russell (AR), Dave Herman (DH) (calling in), Bernd Mathiske (BM), Andreas Rossberg (ARB), Mark Miller (MM), Tom Van Cutsem (TVC), Jasvir Naga (JNA), Istvan Sebestyen (IS)

JN: Going through the agenda Adding proto Unifying iterator/generator APIs Talking about getting user stats for test-262... YK: Prioritize ES6 items. So that we don't get do ES7+ items before

Minutes approved unanimously

link 4.1 Object.freeze

DC: Today Object.freeze throws when primitives are passed in. Suggesting not throwing when a value type is passed in.

MM: Object.isExtensible would return false for primitives

EA: This would give an inconstint view for primitives.

AWB/YK: (In strict mode) numbers and strings lazily box so the assignment never fails.

MM: Proxies are allowed to be non extensible and throw away.

ARB: Is the suggestion to lazily wrap primitives?

MM: No, then Object.isExtensible(7) would return true because the wrapper is extensible.

AWB: In most of the new changes we are not doing unnecessary coercion.

YK: The Chrome dev tools, console.dir(7), says "no properties" which supports treating these as empty objects.

MM: The only observable wrapper is the this wrapper in non strict mode.

AWB: In the new spec, Object.setPrototypeOf(7) throws.

MM: Agrees violently!

link Conclusion/Resolution

• DC+AWB to work out the details

link 4.2 WeakSet

Do we need them?

MM: Trivial shim around WeakMap.

YK: Often wanted it

AWB: Adds no new capabilities.

AR: We should not limit ourselves to what is a new primitive capabilities

AI(AWB): add to spec

link Consensus/Resolution:

• Add WeakSet in ES6

link 4.4 Proxies

TVC's presentation on Notification Proxies: https://docs.google.com/file/d/0B9iYRsLxmdqUd1RsdHZtazliWmc/edit?usp=sharing

Arguments against:

• shifts the burden from spec writers/implementors to users (need to use shadow target even for non-frozen objects)
• implementors will deal with spec bugs related to invariant violations as they come up

link Consensus/Resolution:

• Notification proxies are not approved.
• MM & TVC are still happy with direct proxies.

link Proxy Invoke Trap and wrong |this|-binding on built-in methods

AWB: with current default behavior of "get", "Caretaker" will break on built-ins such as Date, because the |this| binding is by default set to the proxy, so the Date built-in method will not find the correct private state.

ARB: Same issue with binary methods

STH: We should add invoke trap but not change the object model

MM: Pleasant to have. Separate from private state.

AWB: used to think this was an issue with proxies, but convinced that it's an API issue: we need to provide default handlers that do the right thing, and which users can subclass. In particular, want a handler that, on forwarding, rebinds |this| to the target.

STH: If you want to proxy a Date method the underlying this needs to be a non wrapped Date object.

TVC: previously proposed a Handler API that defines derived traps and fundamental traps, allows you to subclass and inherit correct behavior for derived traps. Can be used as the basis.

AWB/TVC: invoke trap would make it easier to control |this|-binding

DH: Never liked breaking the semantics of [[Get]] + [[Call]]

TVC: there already exist invoke-only properties on platforms with __noSuchMethod__

AWB: For a [[Call]] it might be important to control this but by the time the [[Call]] is happening you do not know what this to use.

DH: ActionScript has a proxy and they do have an invoke trap.

BM: The most common action is to invoke a method.

? : we already gave up on the |this| invariant for accessors: in ES5, if obj.x is a getter, |this| will always be bound to obj in the getter. With proxies this is no longer true.

AI(AWB, TVC): Add spec for invoke. Tom and Allen to work out details of a Handler API that accommodates both "caretaker" (aka forwarding) and "virtual object" use cases.

link Consensus/Resolution:

• Add invoke trap.

link 4.11

MM: Everybody in this room wants classes and want to post pone private state to after ES6

ARB: Disagrees.

ARB: Based on feedback, people do not want unique symbols, only private symbols.

MM: Private symbols do not work with proxies.

TVC: can still use WeakMap for private state.

DH: The most common cases where true information hiding is self hosting. The stakes are too high for the browser engines.

YK: If "iterator" would be a private symbol, you cannot create a proxy that will work with for-of loops.

ARB: Symbols (unique and private) and relations overlap.

BE: If we add symbols now we are stuck with them.

LH: Future users will be confused. They will not know what to use

BE: Unique symbol is very different from class private syntax.

AWB/MM: If we first did relationships we might not need symbols.

MM: Relationship published but not reflective.

MM: Difference between relationships and symbols: where is the mutability? This forces us to have both relationships and unique symbols.

link Consensus/Resolution:

• ?

link 4.13 Endianness of Typed array

ARB: Remember it as if we should specify this.

BE: Endianness in Typed Arrays is unspecified.

DH: Keep it open for now... Same system to same system. Using data view, which is explicit, there is no problem.

STH: We don't know what WiiU will do?

AWB: Or they decide not to comply to the spec

DH: WebGL is endian agnostic.

link Consensus/Resolution:

• Leaving it unspecified in ES6.

link 4.18 proto

STH: Recollection, first as data property, then as an accessor. Then discussed the power of that setter. Set the [[Prototype]] in the [[Realm]]. Then Allen wrote the spec. Realized that there were some problems with that design. Roughly the same power as Object.setPrototypeOf.

MM: Existence of a setter... as long as we have the extensibility restriction, that is sufficient.

AWB: Why restrict __proto__ and not other

DH: Objects belonging to a realm is a bad idea.

MM: No more reason to restrict the setter.

STH: Bind __proto__ setter to the object upon extraction

MM: In SES objects that are non extensible. Not going to remove __proto__ going forward.

ARB: If Object.prototype.__proto__ is a data property, making it non writable prevents other objects to use assign to set __proto__.

AWB: If Object.prototype.__proto__ is an accessor that just calls Object.{set,get}PrototypeOf.

AR: Best practice on the web is important even in the future.

TVC: If we have Object.prototype.__proto__ do we want Object.setPrototypeOf or just Reflect.setPrototypeOf?

AWB: Makes sense to have Object.setPrototypeOf for consistency.

EA: Where do we draw the line (Object.x or Reflect.x)?

DH: People will need to be able to get this before we have a reflect module.

TVC: We need both because they have different return value (Reflect.setPrototypeOf returns boolean success value).

link Consensus/Resolution:

• __proto__ is an accessor on Object.prototype.
  • The setter mutates [[Prototype]].
  • There is no "poison pill".
• Add Object.setPrototypeOf and std:reflect setPrototypeOf.

link Naming of @@iterator

AWB: Suffix with $

STH: Opposed to special naming. People don't do this kind of naming convention. Why do we want to introduce this concept?

class Foo {
  *[iterator]() {
    yield ...
  }
}

link Consensus/Resolution:

• No special naming

link Generators and iterators

AWB: send is gone in favor of next(arg) (only first arg is passed through in yield*)

YK: Whether generators return a frozen object or not?

BE: close is gone

link Consensus/Resolution:

• Removed: send and close