Maps don't consume inputs when used in flatten structs?

[Mingun]

I'm wonder of existence of that test:

serde/test_suite/tests/test_annotations.rs

Lines 1928 to 1968 in e6f086d

	fn test_flatten_map_twice() {
	#[derive(Debug, PartialEq, Deserialize)]
	struct Outer {
	#[serde(flatten)]
	first: BTreeMap<String, String>,
	#[serde(flatten)]
	between: Inner,
	#[serde(flatten)]
	second: BTreeMap<String, String>,
	}
	#[derive(Debug, PartialEq, Deserialize)]
	struct Inner {
	y: String,
	}
	assert_de_tokens(
	&Outer {
	first: {
	let mut first = BTreeMap::new();
	first.insert("x".to_owned(), "X".to_owned());
	first.insert("y".to_owned(), "Y".to_owned());
	first
	},
	between: Inner { y: "Y".to_owned() },
	second: {
	let mut second = BTreeMap::new();
	second.insert("x".to_owned(), "X".to_owned());
	second
	},
	},
	&[
	Token::Map { len: None },
	Token::Str("x"),
	Token::Str("X"),
	Token::Str("y"),
	Token::Str("Y"),
	Token::MapEnd,
	],
	);
	}

Why first map don't consume all keys? It seems to me that it should. Current behavior is very unintuitive

[jonasbb]

Have you read the commit message of the commit introducing the code? It explains why. 5ee2fc0

[Mingun]

No, I didn't read, thanks for pointing it. But anyway, this isn't explains why such behaviour considered as correct. Custom struct type interested only in some keys and consumes them from the input. Map<K, V> is a degenerate case of such struct that interested in all keys. I don't understand why it shouldn't consume them.

[dtolnay]

I think my preference would be to stick with the current behavior on this.