Safari forOf and spread issues for Set object.

[patrick-martin-shoutpoint]

When using the Set object, the following code fails in Safari 8 desktop and mobile:
"Error: undefined is not a function (evaluating 'f.next()')"

var set = new Set([1,2,3,4]);

var arr = [...set]; // FAIL

for (var v of set) // FAIL
  console.log(v);

var vals = set.values();
for (var v of vals) // FAIL
  console.log(v);

Note: This code fails to compile in the Traceur REPL on Safari.
https://google.github.io/traceur-compiler/demo/repl.html#var%20set%20%3D%20new%20Set(%5B1%2C2%2C3%2C4%5D)%3B%0A%0Avar%20arr%20%3D%20%5B...set%5D%3B%0A%0Afor%20(var%20v%20of%20set)%0A%20%20console.log(v)%3B%0A%0Avar%20vals%20%3D%20set.values()%3B%0Afor%20(var%20v%20of%20vals)%0A%20%20console.log(v)%3B%0A

spread@https://google.github.io/traceur-compiler/bin/traceur.js:1316:38

getAnonymousModule@https://google.github.io/traceur-compiler/bin/traceur.js:681:34
eval code
eval@[native code]
evaluateCodeUnit@https://google.github.io/traceur-compiler/bin/traceur.js:29800:36
evaluate@https://google.github.io/traceur-compiler/bin/traceur.js:29956:50
evaluate@https://google.github.io/traceur-compiler/bin/traceur.js:30257:37
handleCodeUnitLoaded@https://google.github.io/traceur-compiler/bin/traceur.js:30170:24
module@https://google.github.io/traceur-compiler/bin/traceur.js:30056:32
module@https://google.github.io/traceur-compiler/bin/traceur.js:30317:41
transcode
compileContents
compile
wrappedFunc

[caitp]

yeah, iterator protocol in JSC is still a work in progress. last I checked (in jan/feb), @@iterator wasn't exposed at all in JS, and was only implemented for specific types (just arrays, maybe).

maybe it would make sense to always use the polyfilled version for JSC?

[arv]

We should probably be a lot stricter when it comes to determining when to use the polyfills. One suggestion is to use the polyfills if there is no [Symbol.iterator].

[cwalther]

I just ran into a similar issue with Safari on iOS 8.1.3. The problematic cases in my case were for-of over an Array as well as construction of a Map from an Array (or iterable). The specific reasons were that

• Safari natively implements Array.prototype.entries and Array.prototype.keys, but the iterators returned by them lack a next method.
• Safari natively implements Map, but the constructor does not take an argument.

I can work around the problems by including the following as the first script of the page:

// Remove insufficient built-in implementations so that the polyfills from traceur-runtime kick in.
if (Array.prototype.entries && typeof [1].entries().next !== 'function') {
    delete Array.prototype.entries;
}
if (Array.prototype.keys && typeof [1].keys().next !== 'function') {
    delete Array.prototype.keys;
}
try {
    new Map([]);
}
catch(e) {
    delete Map;
}

I assume these checks could be built into the polyfillArray and polyfillMap functions but have not tried that.

[arv]

Safari natively implements Array.prototype.entries and Array.prototype.keys, but the iterators returned by them lack a next method.

WHAT!! How do you iterate over them then? How can they ship something so broken... makes me very sad.

@cwalther What does Map.length return in Safari? If it says 0 we should be able to use that instead of using the try/catch.

[cwalther]

I have no idea what the purpose of these iterators is, they appear to have no methods (or other properties) at all besides the ones inherited from Object.

Map.length indeed returns 0. (I wasn’t even aware that functions had a length property, thanks for pointing me to that!)

Interestingly, Set.length is 0 as well, but new Set([4, 5]) still works (Set([4, 5]) on the other hand throws TypeError: Set does not accept arguments when called as a function).

[arv]

TypeError: Set does not accept arguments when called as a function

Sad. So misleading. Set throws when called.

https://people.mozilla.org/~jorendorff/es6-draft.html#sec-set-iterable

Step 1

[cwalther]

I just see in that document that Map.length should be 0, unless I’m misunderstanding something, so maybe checking for that is not a suitable way of detecting insufficient implementations after all.

By the way, es6-shim also fixes the issues I had on that version of Safari, and it has a whole bunch of criteria for checking whether to replace Map, maybe that’s a more reliable source. (I didn’t use es6-shim as a final solution because I was wary of potential unintended interactions between es6-shim and traceur-runtime.)

[arv]

You are correct. Map.length should be 0.

I think a suitable test is to see if Map.prototype[Symbol.iterator] is a function.

[zloirock]

@cwalther @arv bad idea. es6-shim also has a problem - it's replace collections in all modern environment. Maybe it's partly fixed after yesterday's reworking. My point about collections polyfilling here.