Patch arrays generated after removing elements differs about json-patch

[Starcounter-Jack]

Thanks for reporting this. When it comes to the order for the patches in the shim, they should be reversed (as you pointed out).

When it comes to the issue of the length property, I'm not sure which way to go. length is not an enumerable property, so should we generate a patch for it? As it is ignored by JSON.stringify(), the default behaviour should perhaps be that it is ignored by the JSON-Patch generator.

WDYT?

[warpech]

The length property has it's own ticket: #14. I think it should be removed from the output

[warpech]

Native Object.observe is reporting items removed with Array.prototype.splice in reverse order. In contrary, the shimmed version reports removed items in the natural order.

As we see here, different implementations may produce different results depending on the direction of the algorithm. When more browsers support Object.observe, that may get even more diverse if you compare the results as strings.

To me, the only logical way to sort patches is by the time of detection. Because we have no control over order of changes made internally in Array.prototype.splice, we need to agree that this may be reverse order if the browser implements it so.

Native Object.observe may know the real order of changes, but shim may only traverse the tree on time intervals, and then report the changes in natural (alphabetical) order.

Thus, in my opinion this issue should be Won't fix, but I leave this open for discussion.

[warpech]

Now I have a new light on this issue... Considering this array:

var obj = {
  items: ["a", "b"]
}

Calling this will work OK:

jsonpatch.apply(obj, [
  {
    "op": "remove",
    "path": "/items/1"
  },
  {
    "op": "remove",
    "path": "/items/0"
  }  
])

But calling this will call exception:

jsonpatch.apply(obj, [
  {
    "op": "remove",
    "path": "/items/0"
  },
  {
    "op": "remove",
    "path": "/items/1"
  }
])

Because after first patch is applied, index 1 no longer exists in the items array. I think we need to change the shim implementation to iterate backwards so the array changes are always reported in descending index order.

[warpech]

Fixed in 0.3.4