How to merge objects in a collection with the same key

[jwanggg]

Hi I was looking at the object comprehension in order to decompose objects and transform them but am not getting a correct result.

Given my data

[{
		"input1": [
			"a",
			"b"
		]
	},
	{
		"input1": [
			"a",
			"c",
			"d"
		]
	},
	{
		"input2": [
			"a"
		]
	}
]

How would I transform it into just

[{
		"input1": [
			"a",
			"b",
			"c",
			"d"
		]
	},
	{
		"input2": [
			"a"
		]
	}
]

I tried with a more complex dataset and query here https://play.openpolicyagent.org/p/Pw6kSPNU4R. Not sure why it doesn't merge together when the keys are the same as well

[johanfylling]

Hi Jonathan!

I'm a bit rusty, so there are likely more elegant solutions, but this should do the trick:

package play

merged[x] {                            # 4. putting all objects in a set
    input[_][key]                      # 1. enumerating the unique keys
    set := {v | v := input[_][key][_]} # 2. joining all list items with the same key into a set
    x := {key: set}                    # 3. making an object out of the key and associated set
}

https://play.openpolicyagent.org/p/H5fHwLc6yR

[jwanggg]

Thanks @johanfylling I tried with a more complex data sample and input here https://play.openpolicyagent.org/p/Pw6kSPNU4R. Seems like they don't merge correctly. I wonder if it's because it is too nested? Trying to merge the first application keys and then merge the experiment part as well so it's like twice

[johanfylling]

I don't quite follow what you're trying to accomplish. Could you please provide a sample of your expected output?

[jwanggg]

Sorry for the late reply, essentially the above example I wanted the two experiment keys merged. Here's an updated policy where the permissions in the output is what I wanted https://play.openpolicyagent.org/p/jMmZmU6zTX. Permissions0 and Permissions1 are the steps before merging into getting it into the final Permissions. Wondering if there was a way to simplify everything.

[anderseknert]

For simple merges you have the object.union and object.union_n built-in functions to work with. There was a recent proposal by @charlesdaniels to add another one with object.union_merge, but as your requirements show, there aren't really any agreed-upon resolutions for merging values like arrays... so we'd either need to have one built-in per option, or provide some generic way of bringing in your own "conflict resolver".. which would be a challenge in a non-functional language like Rego, I think.

So for the time being, you're probably best off with a custom implemnentation like the one @johanfylling provided here.