Optional maps get created when `raw_value == nil`

[bmurphy23]

If a map attribute is marked as optional, and the incoming params do not have that data, then the struct that Speck outputs will have an empty(-ish) map for that attribute. Here's a barebones example:

struct MyNamespace.FooData

name "foo_data"

attribute :name, :string
attribute :status, :atom, values: [:ok, :not_okay]

attribute :data, optional: true do
  attribute :timestamp, :datetime
  attribute [:values] do
    attribute :value_1, :integer
    attribute :value_2, :string
  end
end

Then, in iex:

json = """
{
  "name": "foo",
  "status": "ok"
}
"""
message = Jason.decode!(json)

Speck.validate(MyNamespace.FooData, message)

If you run this, you'll get this returned:

{
  :ok,
  %MyNamespace.FooData{
    name: "foo",
    status: :ok,
    data: %{
      values: []
    }
  }
}

There are two interesting things I'm seeing here:

• :data exists at all in the returned struct.
• :values is included in the map, but :timestamp isn't.

[bmurphy23]

After reading some more code, I realize that this example isn't fully demonstrative. Since Speck has to defstruct using the top level attributes, those must always be present in the returned struct. However, the same behaviour as noted above occurs if the optional map is not top-level, like so:

struct NameSpace.DoFoo

name "do_foo"

attribute :top_level do
  attribute :inner_level, optional: true do
    attribute :value, :integer
    attribute :name, :string
  end
end

If the params given to Speck.validate don't include anything for :inner_level, the map will still exist in the returned struct.

[amclain]

Thanks Ben!

I believe this is happening because attribute [:values] do defaults to [] instead of nil, which on its own is good, but when combined with the parent optional: true, it's a problem. The expected value should be data: nil.

I have a feeling that attribute [:values], optional: true do is a quick workaround. I unfortunately don't have time to check it right now.

We're pretty booked up right now, so we probably won't be able to slot this into our free support tier until after NervesConf next month. However, we could review a PR if you're able to come up with the fix before then.

[amclain]

I saw this issue pop up with a list of maps recently, and yeah, it's an issue. The result looks like this:

%{
  "attribute_1" => [
    %{},
    %{"attribute_2" => 2}
  ]
}

It's captured (incorrectly) in a unit test that's landing in main soon, so removing the empty map from that test is the starting point for this fix.

speck/test/speck_test.exs

Line 704 in c5ba1ee

%{},