How to represent BitFlags?

[ascjones]

In substrate we have the following type:

#[repr(u64)]
#[derive(Encode, Decode, Clone, Copy, PartialEq, Eq, BitFlags, RuntimeDebug, TypeInfo)]
pub enum IdentityField {
	Display = 0b0000000000000000000000000000000000000000000000000000000000000001,
	Legal = 0b0000000000000000000000000000000000000000000000000000000000000010,
	Web = 0b0000000000000000000000000000000000000000000000000000000000000100,
	Riot = 0b0000000000000000000000000000000000000000000000000000000000001000,
	Email = 0b0000000000000000000000000000000000000000000000000000000000010000,
	PgpFingerprint = 0b0000000000000000000000000000000000000000000000000000000000100000,
	Image = 0b0000000000000000000000000000000000000000000000000000000001000000,
	Twitter = 0b0000000000000000000000000000000000000000000000000000000010000000,
}

It is wrapped in a BitFlags in the following wrapper type to allow custom encoding/decoding.

/// Wrapper type for `BitFlags<IdentityField>` that implements `Codec`.
#[derive(Clone, Copy, PartialEq, Default, RuntimeDebug)]
pub struct IdentityFields(pub(crate) BitFlags<IdentityField>);

Question is how to represent this with scale_info: it is encoded as a raw u64 but we need to also have access to the IdentityField enum definition which maps the discriminants reduced to u8 indices.

A temporary workaround is to provide the following TypeInfo implementation

impl TypeInfo for IdentityFields {
	type Identity = Self;

	fn type_info() -> Type {
		Type::builder()
			.path(Path::new("BitFlags", module_path!()))
			.type_params(vec![ TypeParameter::new("T", Some(meta_type::<IdentityField>())) ])
			.composite(
			Fields::unnamed()
				.field(|f| f.ty::<u64>().type_name("IdentityField")),
		)
	}
}

See how it represents BitFlags<IdentityField> including the IdentityField as a type parameter, and the field itself as type u64.

However this requires some string matching for downstream type generators, so a preferable solution might be to add fist class support for a BitFlags like TypeDef variant.

On the other we might not want to be adding a new TypeDef variant for each corner case, be curious to see what others think.

/cc @jacogr

[ascjones]

Another possibility is to capture the #[repr(u64)] and encode that into an optional repr field in TypeDefVariant

[gui1117]

sidenote: the Encode/Decode implementation on IdentityField is wrong and should be removed paritytech/substrate#9445

[ascjones]

Then should I reconsider the removal of discriminant: u64 in #112?

[gui1117]

if the only use of discriminant is bitflag maybe it is better to have dedicated syntax.

But otherwise it is true that the discriminant gives useful information in this specific case.