Substitution can break overrides

[JettJones]

I expect that overrides would replace the previous values when merging objects. In this example that works only some of the time - possibly depending on how many variable overrides there are?

{
    name: WRONG
    new_name: right
    value: two

    test_set: {
        test1_long:         something-${value}-${name}-suffix
        test2_long_str:     something-${value}-${name}-suffix
        test3_short:        something-${value}-${name}-suffix
        test4_short_str:    something-${value}-${name}-suffix
    }

    test_set: {
        test1_long:         something-${value}-${new_name}-suffix
        test2_long_str:     something-${value}-legacy-suffix
        test3_short:        x-${new_name}-suffix
        test4_short_str:    x-legacy-suffix
    }
}

Expected:

{
	...
    test_set: {
        test1_long:         something-two-right-suffix
        test2_long_str:     something-two-legacy-suffix
        test3_short:        x-right-suffix
        test4_short_str:    x-legacy-suffix
    }
}

Observed:

{
	...
    test_set: {
        test1_long:         something-two-right-suffix
     	test2_long_str: 	something-two-WRONG-suffix
		test3_short: 		something-two-WRONG-suffix
        test4_short_str:    x-legacy-suffix
    }
}

From a quick code dive, I think this is happening during resolve_substitutions - in order to support self-referential substitutions, overridden substitutions are still evaluated, and they are applied to the resulting object. Depending on if the shadowed fields are evaluated before the final values, the wrong result can be left in the final config object.