[v0ropaev]

When a proto2 required field (or an editions LEGACY_REQUIRED one) declares [default = X], the owned struct gets a hand-written Default that starts the field at X, so Message::decode of input without the field gives X back. The eager and lazy views derived Default instead, unless an enum-type override had opened a required enum, so the same field held the type's zero value. For the same bytes the view, its re-encoding and to_owned_message() then disagreed with owned decode. With the existing RequiredDefaults fixture:

let owned = RequiredDefaults::decode_from_slice(&[]).unwrap();
let view = RequiredDefaultsView::decode_view(&[]).unwrap();
assert_eq!(owned.count, 42);
assert_eq!(view.count, 42); // was 0
assert_eq!(view.level, Priority::CRITICAL); // was LOW
// view.encode_to_vec() wrote 08 00 where the owned encoding writes 08 2a

C++ after ParsePartial and Go after an AllowPartial unmarshal also return the declared default from their getters.

custom_view_default_impl now fills these fields from their declared default, under the same gate the owned Default uses (explicit presence on the message, or an opened enum). View strings and bytes borrow, so they get a 'static literal, and a string that strict_utf8_mapping turned into bytes keeps its text as UTF-8. Every other type reuses parse_default_value, and the opened-enum path is unchanged. The lazy view picks this up through the same function.

The new tests check the view values for an empty buffer and a partial one, and check that to_owned_message() and the view's own encoding match owned decode. They cover RequiredDefaults, RequiredBytesDefaults with its nested message in the three bytes configurations, KeywordEnumDefault on the lazy view, and the editions LEGACY_REQUIRED fixture in utf8_validation.proto. Checked-in generated code doesn't change.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}