Chrome re-ordering object keys if numerics, is that normal/expected

11,162

Solution 1

It's the way v8 handles associative arrays. A known issue Issue 164 but it follows the spec so is marked 'working as intended'. There isn't a required order for looping through associative arrays.

A simple workaround is to precede number values with letters e.g: 'size_7':['9149','9139'] etc.

The standard will change in the next ECMAScript spec forcing [chrome] developers to change this.

Solution 2

It would appear that Chrome is treating the integer string as if it were a numeric type when used as an index/property name.

I think relying on the Javascript implementation to preserve the order of what, in some cases, is object properties, and in other cases (certainly with chrome) array indices, is demonstrably an unsafe approach and order of enumeration is probably not defined in the spec. I would suggest adding an additional property to the JSON that indicates a sort order:

{
    "7":{"sortOrder":1,"data":["9149","9139","10455","17208"]},
    "7.5":{"sortOrder":2,"data":["9140","9150","10456","17209"]}
    //etc
}

Solution 3

They're being treated as strings because they are strings. My best suggestion would be to use the same "precision" in all your keys.

{
  "7.0": ["9149", "9139", "10455", "17208"],
  "7.5": ["9140", "9150", "10456", "17209"],
  "8.0": ["2684", "9141", "10457", "17210"],
  "8.5": ["9142", "10444", "10458", "17211"],
  "9.0": ["2685", "9143", "10459", "17212"],
  "9.5": ["10443", "9144", "10460", "17213"]
}

So, "8.0" instead of "8", etc.

Even then, there are no guarantees, but it is more likely that they'll come out in the same order.

For a better guarantee, perform a sort based on the keys, putting the values into an array in the sorted order.

Solution 4

When iterating over the properties of an object, the order is specified in the ECMAScript specification as being undefined and any order you may have observed in some environment should not be relied upon. If you need order, use an Array.

Share:
11,162
Dimitar Christoff
Author by

Dimitar Christoff

Javascript is to java what hamster is to ham. Follow me on Github

Updated on June 25, 2022

Comments

  • Dimitar Christoff
    Dimitar Christoff almost 2 years

    I noticed that certain code that evaluates some shoe sizes for an e-commerce site and outputs them on screen is messing up the order in Chrome.

    JSON given can be:

    {
      "7": ["9149", "9139", "10455", "17208"],
      "7.5": ["9140", "9150", "10456", "17209"],
      "8": ["2684", "9141", "10457", "17210"],
      "8.5": ["9142", "10444", "10458", "17211"],
      "9": ["2685", "9143", "10459", "17212"],
      "9.5": ["10443", "9144", "10460", "17213"]
    }
    

    ...which increments sizes in halves.

    Upon conversion into an object and iterating though the keys, the natural order is being respected and they come out as:

    7, 7.5, 8, 8.5 etc.

    But in Chrome only, keys that 'look' like round numbers ALWAYS come out of the object first, so output of a for... in loop is:

    7, 8, 9, 7.5, 8.5, 9.5 ...

    Object.keys(sizes); // ["7", "8", "9", "7.5", "8.5", "9.5"]
    

    Here is the test case: https://jsfiddle.net/wcapc46L/1/

    It's only affects whole numbers, seems like Webkit / Blink have an optimisation that prefers Object properties that are numeric, maybe it's to do with Branch Prediction or whatever.

    If you prefix the object keys with any character, the order remains unaffected and works as intended - FIFO

    I think I recall reading that there are no guarantees on the order of properties of an object but at the same time, this is annoying to the extreme and would cause a considerable amount of effort in fixing it for chrome users alone.

    Any ideas? is this likely a bug that will get fixed?

    edit additionally, I have now discovered this as an issue on the v8 bug tracker:

    https://code.google.com/p/v8/issues/detail?id=164

    Looks like Blink don't want to fix this and will remain the only browser that will do it.

    update whatever hash table optimisation webkit/blink had, has now made its way into gecko (FF 27.0.1) - https://jsfiddle.net/9Htmq/ results in 7,8,9,7.5,8.5,9.5. applying _ before the keys returns the correct / expected order.

    update 2017 People are still upvoting and editing this so - It does NOT appear to affect Map / WeakMap, Set etc (as demonstrated by updated main example)

  • Dimitar Christoff
    Dimitar Christoff almost 14 years
    change in favour of chrome's current implementation or in favour of preserving the order of definition?
  • Dimitar Christoff
    Dimitar Christoff almost 14 years
    regretfully, even though this will work for the test case, this won't be an option for the implementation. sizes are a multitude of various options, such as `S,M,L,XL,XXL' or '44,44.5,44.5' or 'One Size' and so forth, as defined on a per-product basis by the merchant. there is tremendous logic involved in sorting the output from MYSQL so it arrives in the correct order and not as strings - it covers something like 40 separate sizes pre-defined in order of preference.
  • Michael Sparks
    Michael Sparks almost 14 years
    preserving the order since all other browsers already do this.
  • Tim Down
    Tim Down almost 14 years
    "The standard will change in the next ECMAScript spec forcing developers to change this." Really? Where's the evidence for this? The recent ECMAScript 5 spec doesn't specify an order. Sounds like it got dropped at the last minute; who's to say that won't happen again next time?
  • Dimitar Christoff
    Dimitar Christoff almost 14 years
    perhaps implied for harmony and not ECMA 5
  • Michael Sparks
    Michael Sparks almost 14 years
    It might not be part of the spec. I read about it in John Resig's blog . ejohn.org/blog/javascript-in-chrome
  • Dimitar Christoff
    Dimitar Christoff almost 14 years
    hrm he does state All modern implementations of ECMAScript iterate through object properties in the order in which they were defined. Because of this the Chrome team has deemed this to be a bug and will be fixing it. - i read that as a positive sign the--again it got posted in 2008 and it's still not the case.
  • Tim Down
    Tim Down almost 14 years
    Well, silly Mr Resig, if he relied on this. Relying on non-uniformly observed rather than specified behaviour is a recipe for bad, unreliable code.
  • Dimitar Christoff
    Dimitar Christoff almost 14 years
    that can work - i can apply a filter that checks sortOrder property upon looping and injects the elements accordingly in the DOM.
  • gsnedders
    gsnedders about 11 years
    Chakra and Carakan (in IE9+ and Opera 10.50+) also have this behaviour, not just V8.
  • Reality-Torrent
    Reality-Torrent over 8 years
    I tried turning the index to a string instead of a number, but Chrome seems to rearrange the order anyway. @Michael Sparks
  • Reality-Torrent
    Reality-Torrent over 8 years
    And that is not an answer.
  • Michael Sparks
    Michael Sparks over 8 years
    @Reality-Torrent Simply converting to a string isn't enough here. You'll need a character before the key. Here is an example. jsfiddle.net/9Htmq/5 A good way to think about it is, "What will this return when passed to parseInt?". If the answer is NaN it should work.
  • Ben Aston
    Ben Aston about 4 years
    @MichaelSparks With ES2020, I think this answer is now incorrect.