rmp-serde

chore: Release rmp-serde version 1.3.0

chore: Release rmp-serde version 1.2.0

(cargo-release) rmp-serde version 1.1.1

Changed:

• Raw and RawRef are now serializable.
• Allow to construct Raw and RawRef from string or from a byte array.

Added:

• Serialize struct as a map (#140).

Changed

• Switch to using char::encode_utf8. In Rust 1.15, the function char::encode_utf8 was stabilized. Assuming that rmp follows the serde standard of supporting the last 3 stable releases, this function is now safe to use. I believe this removes the last allocation required on the serialization path.

Fixed

• Fixed build on nightly rustc (#135).

Fixed

• Fixed build on nightly rustc (#131).

Added

• Add helper RawRef struct that allows to deserialize borrowed strings even if they contain invalid UTF-8. This can be when deserializing frames from older MessagePack spec.

Added

• Zero-copy deserialization from &[u8].

Changed

• Adapt with serde 1.0.

Fixed

• Fix compilation on rustc 1.13.

Added

• Add helper Raw struct that allows to deserialize strings even if they contain invalid UTF-8. This can be when deserializing frames from older MessagePack spec.
• Serializer can now return back its underlying writer by reference, mutable reference and by value.

Added

• Added write, to_vec and from_read functions to reduce boilerplate for serializing and deserializing custom types that implement Serialize or Deserialize.

Changed

• Adapt with serde 0.9.

Changed

• Adapt with RMP core 0.8.
• The Serializer now encodes integers using the most effective representation.
• The Deserializer now properly decodes integer values that fit in the expected type.
• Default stack protector depth is now 1024 instead of 1000.
• Internal buffer in the Deserializer now have some capacity preallocated.

Changed

• Update serde dependency to 0.8.

Fixed

• Switch unit structs to using the same serialization mechanism as other structs (#76).

Fixed

• Reading binary should no longer trigger unexpected EOF error on valid read.

Changed

• Reuse deserializer buffer on every read for string and binary deserialization without unnecessary intermediate buffer creation. This change increases the string and binary deserialization performance (many thanks to Fedor Gogolev [email protected]).

Added

• Implement size_hint() function for SeqVisitor and MapVisitor, so it can be possible to preallocate things, increasing the performance greatly.

Fixed

• Serializer should no longer panic with unimplemented error on struct variant serialization ([#64]).

Changed

• Adapt code to be compilable with Serde v0.7.

Changed

• Fixed stack overflow when unpacking recursive data structures.

Changed

• Upper limit for serde version.

Fixed

• Use the most effective int encoding Even if the value is explicitly marked as i64 it must be encoded using the most effective bytes representation despite of signed it or unsigned.

Changed

• Serializer can now be extended with custom struct encoding policy.
• Improved error types and its messages for serialization part.
  • New error type introduced - UnknownLength. Returned on attempt to serialize struct, map or serquence with unknown length (Serde allows this).
  • The new type is returned if necessary.

Fixed

• Deserializer now properly works with enums.
• Options with default values (that can be initialized using unit marker) deserialization. This fix also forbids the following Option deserialization cases:
  • Option<()>.
  • Option<Option<...>>. It's impossible to properly deserialize the listed cases without explicit option marker in protocol.
• Serializer now properly serializes unit structs. Previously it was serialized as a unit (nil), now there is just an empty array ([]).