Unexpected behavior with walk path as key in object comprehension

[robmyersrobmyers]

Description

I'm seeing some unexpected behavior when usingpath from walk as an object key inside a comprehension.
It's a bit strange to use an array as a key, so I'm not totally sure if this is a bug. However, it seems to work reliably when not using walk in the object comprehension.

I verified you can construct an object with arrays as keys:

	a := ["a", 0, "b", 0, "c", 0, "aa"]
	b := ["a", 0, "b", 0, "c", 1, "bb"]
	c := ["a", 0, "b", 0, "c", 2, "cc"]
	obj := {
		a: "AA",
		b: "BB",
		c: "CC",
	}

	print(obj)

results in {["a", 0, "b", 0, "c", 0, "aa"]: "AA", ["a", 0, "b", 0, "c", 1, "bb"]: "BB", ["a", 0, "b", 0, "c", 2, "cc"]: "CC"}

But I ran into issues using the path array from the walk() builtin to do something similar.

x := {"a": [{"b": [{"c": [
{"aa": "AA"},
{"bb": "BB"},
{"cc": "CC"},
{"dd": "DDD"},
]}]}]}
result := {path: value |
[path, value] := walk(x)
count(value) == 2
}

print(result)

results in {["a", 0, "b", 0, "c", 3, "dd"]: "AA", ["a", 0, "b", 0, "c", 3, "dd"]: "BB", ["a", 0, "b", 0, "c", 3, "dd"]: "CC"}, which is almost correct, but the key is repeated thrice instead of corresponding to the value's path.

I ran a similar test manually creating the object with a comprehension:

	a := [["a", 0, "b", 0, "c", 0, "aa"], "AA"]
	b := [["a", 0, "b", 0, "c", 1, "bb"], "BB"]
	c := [["a", 0, "b", 0, "c", 2, "cc"], "CC"]
	d := [a, b, c]
	obj := {item[0]: item[1] |
		some item in d
	}

	print(obj)

which seems to work and resulted in {["a", 0, "b", 0, "c", 0, "aa"]: "AA", ["a", 0, "b", 0, "c", 1, "bb"]: "BB", ["a", 0, "b", 0, "c", 2, "cc"]: "CC"}

• OPA version : 1.5.0

Expected behavior

I expected the object comprehension using the walk builtin to work the same way it did when created manually.

Additional context

If this is not a bug, then some additional documentation and additional examples might help others avoid this pitfall.

[anderseknert]

Thanks for reporting this @robmyersrobmyers. A detailed report as always ⭐

Looks like this is a regression in v1.5.0, no doubt introduced in #7612 by yours truly. Sorry about that!

The walk built-in has a fairly decent coverage, in the tests here to some extent, but an extremely good coverage in Regal, where we likely walk more than what's done in any other project. So I thought we'd be good, but clearly you've found a curious case where not deep-copying every path element in each iteration breaks things. I can't say I have the slightest idea about what makes this particular case special, but let's try to find out! It would be a shame to roll back the change entirely, but obviously that's what we'll have to do if it comes to that.