2022-07-01 Unreleased version
# C++
* cpp_generated_lib_linked support is removed in protoc
* Reduced .pb.o object file size slightly by explicitly instantiating
  InternalMetadata templates in the runtime.
* Add C++20 keywords guarded by PROTOBUF_FUTURE_CPP20_KEYWORDS
* Fixed crash in ThreadLocalStorage for pre-C++17 compilers on 32-bit ARM.
* Clarified that JSON API non-OK statuses are not a stable API.
* Added a default implementation of MessageDifferencer::Reporter methods.
* proto2::MapPair is now an alias to std::pair.
* Hide C++ RepeatedField::UnsafeArenaSwap
* Use table-driven parser for reflection based objects.
* Update Map's InternalSwap() to take a pointer to the other Map.
* Add ARM-optimized Varint decoding functions.
* Minor optimization for parsing groups
* Declare ReflectiveProtoHook class
* Reduce size of VarintParse code in protocol buffers, by calling the shared
  routine after handling just one-byte varint encoding inline, rather than
  handling one-byte and two-byte varints inline.
* Avoid inlining some large heavily duplicated routines in repeated_ptr_field.h
* Add ReflectiveProtoHook to Reflection.
* Turns on table-driven parser for reflection based objects.
* Save code space by avoiding inlining of large-in-aggregate code-space MessageLite::~MessageLite destructor.
* Undefine the macro `linux` when compiling protobuf
* Reduce memory consumption of MessageSet parsing.
* Save code space by avoiding inlining of large-in-aggregate code-space MessageLite::~MessageLite destructor.
* Breaking change: delete Arena::Init
* Make a PROTOBUF_POISON/UNPOISON to reduce noise in the source
* Put alignment functions in "arena_align.h"
* Split off `cleanup` arena functions into "arena_cleanup.h"
* Fix signed / unsigned match in CHECK_EQ
* Kill Atomic<>. it's not pulling it's weight
* Move AllocationPolicy out of arena_impl, and unify arena_config for bazel
* Fix failure case in table-driven parser.
* Add a new JSON parser.
* Removed old JSON parsing code.
* Introduce the Printer::{SetRedactDebugString,SetRandomizeDebugString} private flags.
* Introduce global flags to control Printer::{SetRedactDebugString, SetRandomizeDebugString}.
* proto3 string fields no longer trigger clang-tidy warning bugprone-branch-clone.
* Fix the API of DescriptorUpgrader::set_allow_unknown_dependencies to set to True always, and to populate into the DescriptorPool as well.
* Report line numbers consistently in text-format deprecated-field warnings.
* Reserve C++20 keywords
* Fixed C++ code generation for protos that use int32_t, uint32_t, int64_t, uint64_t, size_t as field names.
* Annotate generated C++ public aliases for enum types.
* Change default arena max block size from 8K to 32K.

# Kotlin
* Suppress deprecation warnings in Kotlin generated code.
* Kotlin generated code comments now use kdoc format instead of javadoc.
* Escape keywords in package names in proto generated code
* Add Kotlin enum int value getters and setters

# Java
* Performance improvement for repeated use of FieldMaskUtil#merge by caching
  constructed FieldMaskTrees.
* Optimized Java proto serialization gencode for protos having many extension ranges with few fields in between.
* More thoroughly annotate public generated code in Java lite protocol buffers.
* Fixed Bug in proto3 java lite repeated enum fields. Failed to call copyOnWrite before modifying previously built message. Causes modification to already "built" messages that should be immutable.
* Fix Java reflection serialization of empty packed fields.
* Refactoring java full runtime to reuse sub-message builders and prepare to migrate parsing logic from parse constructor to builder.
* Fix TextFormat parser to build up recurring (but supposedly not repeated) sub-messages directly from text rather than building a new sub-message and merging the fully formed message into the existing field.
* Make message-type extensions merge from wire-format instead of building up instances and merging afterwards. This has much better performance.
* Change the Lite runtime to prefer merging from the wireformat into mutable messages rather than building up a new immutable object before merging. This way results in fewer allocations and copy operations.
* Move proto wireformat parsing functionality from the private "parsing constructor" to the Builder class.
* Refactoring java full runtime to reuse sub-message builders and prepare to migrate parsing logic from parse constructor to builder.
* Move proto wireformat parsing functionality from the private "parsing constructor" to the Builder class.
* Change the Lite runtime to prefer merging from the wireformat into mutable messages rather than building up a new immutable object before merging. This way results in fewer allocations and copy operations.
* Make message-type extensions merge from wire-format instead of building up instances and merging afterwards. This has much better performance.
* Fix TextFormat parser to build up recurring (but supposedly not repeated) sub-messages directly from text rather than building a new sub-message and merging the fully formed message into the existing field.
* Fix bug in nested builder caching logic where cleared sub-field builders would remain dirty after a clear and build in a parent layer. https://github.com/protocolbuffers/protobuf/issues/10624
* Performance test for TextFormat to verify fix for https://github.com/protocolbuffers/protobuf/security/advisories/GHSA-h4h5-3hr4-j3g2

# Python
* Changes ordering of printed fields in .pyi files from lexicographic to the same ordering found in the proto descriptor.
* Adds GeneratedCodeInfo annotations to python proto .pyi outputs as a base64 encoded docstring in the last line of the .pyi file for code analysis tools.
* Fix message factory's behavior in python cpp extension to return same message classes for same descriptor, even if the factories are different.
* Add type annotation for enum value fields in enum classes.

# Compiler
* Print full path name of source .proto file on error
* Include proto message type in the annotation comments.