Protobuf

C++
* Optimize and simplify implementation of RepeatedPtrFieldBase
* Don't create unnecessary unknown field sets.
* Remove branch from accessors to repeated field element array.
* Added delimited parse and serialize util.
* Reduce size by not emitting constants for fieldnumbers
* Fix a bug when comparing finite and infinite field values with explicit tolerances.
* TextFormat::Parser should use a custom Finder to look up extensions by number if one is provided.
* Add MessageLite::Utf8DebugString() to make MessageLite more compatible with Message.
* Fail fast for better performance in DescriptorPool::FindExtensionByNumber() if descriptor has no defined extensions.
* Adding the file name to help debug colliding extensions
* Added FieldDescriptor::PrintableNameForExtension() and DescriptorPool::FindExtensionByPrintableName().
The latter will replace Reflection::FindKnownExtensionByName().
* Replace NULL with nullptr
* Created a new Add method in repeated field that allows adding a range of elements all at once.
* Enabled enum name-to-value mapping functions for C++ lite
* Avoid dynamic initialization in descriptor.proto generated code
* Move stream functions to MessageLite from Message.
* Move all zero_copy_stream functionality to io_lite.
* Do not create array of matched fields for simple repeated fields
* Enabling silent mode by default to reduce make compilation noise. (6237)

Java
* Expose TextFormat.Printer and make it configurable. Deprecate the static methods.
* Library for constructing google.protobuf.Struct and google.protobuf.Value
* Make OneofDescriptor extend GenericDescriptor.
* Expose streamingness of service methods from MethodDescriptor.
* Fix a bug where TextFormat fails to parse Any filed with > 1 embedded message sub-fields.
* Establish consistent JsonFormat behavior for nulls in oneofs, regardless of order.
* Update GSON version to 3.8.5. (6268)
* Add `protobuf_java_lite` Bazel target. (6177)

Python
* Change implementation of Name() for enums that allow aliases in proto2 in Python
to be in line with claims in C++ implementation (to return first value).
* Explicitly say what field cannot be set when the new value fails a type check.
* Duplicate register in descriptor pool will raise errors
* Add __slots__ to all well_known_types classes, custom attributes are not allowed anymore.
* text_format only present 8 valid digits for float fields by default

JavaScript
* Add Oneof enum to the list of goog.provide

PHP
* Rename get/setXXXValue to get/setXXXWrapper. (6295)

Ruby
* Remove to_hash methods. (6166)

C++
* Use std::atomic<int32> in case of myriad2 platform
* Always declare enums to be int-sized
* Added DebugString() and ShortDebugString() methods on MessageLite
* Specialized different parse loop control flows
* Make hasbits potentially in register. The or's start forming an obstacle because it's a read modify store on the same mem address on each iteration.
* Move to an internal MACRO for parser validity checks.
* Improve map parsing performance.
* Make MergePartialFromCodedStream non virtual. This allows direct calls, potential inlining and is also a code health improvement
* Add an overall limit to parse_context to prevent reading past it. This allows to remove a annoying level of indirection.
* Fix a mistake, we shouldn't verify map key/value strings for utf8 in opt mode for proto2.
* Further improvements to cut binary size.
* Prepare to make MergePartialFromCodedStream non-virtual.
* A report on some interesting behavior change in python (caused by b/27494216) made me realize there is a check that needs to be done in case the parse ended on a end group tag.
* Add a note of caution to the comments around skip in CodedOutputStream.
* Simplify end check.
* Add overload for ParseMessage for MessageLite/Message types. If the explicit type is not known inlining won't help de-virtualizing the virtual call.
* Reduce linker input. It turns out that ParseMessage is not inlined, producing template instantiations that are used only once and save nothing but cost more.
* Improve the parser.
* [c++17] Changed proto2::RepeatedPtrField iterators to no longer derive from the deprecated std::iterator class.
* Change the default value of case_insensitive_enum_parsing to false for JsonStringToMessage.
* Add a warning if a field name doesn't match the style guide.
* Fix TextFormat not round-trip correctly when float value is max float.
* Added locationed info for some errors at compiler
* Python reserved keywords are now working with getattr()/setattr() for most descriptors.
* Added AllowUnknownField() in text_format
* Append '_' to C++ reserved keywords for message, enum, extension
* Fix MSVC warning C4244 in protobuf's parse_context.h.
* Updating Iterators to be compatible with C++17 in MSVC.
* Use capability annotation in mutex.h
* Fix "UndefinedBehaviorSanitizer: cfi-bad-type"
* CriticalSectionLock class as a lightweight replacement for std::mutex on Windows platforms.
* Removed vestigial wire_format_lite_inl.h

C
* Added System.Memory dependency.

Java
* Make Java protoc code generator ignore optimize_for LITE_RUNTIME. Users should instead use the Java lite protoc plugin.
* Change Extension getMessageDefaultInstance() to return Message instead of MessageLite.
* Prevent malicious input streams from leaking buffers for ByteString or ByteBuffer parsing.
* Release new Javalite runtime.
* Show warning in case potential file name conflict.
* Allow Java reserved keywords to be used in extensions.
* Added setAllowUnknownFields() in text format
* Add memoization to ExtensionRegistryLite.getEmptyRegistry()
* Improve performance of CodedOutputStream.writeUInt32NoTag
* Add an optimized mismatch-finding algorithm to UnsafeUtil.
* When serializing uint32 varints, check that we have MAX_VARINT32_SIZE bytes left, not just MAX_VARINT_SIZE.
* Minor optimization to RopeByteString.PieceIterator

JavaScript
* Simplify generated toObject code when the default value is used.

Python
* Changes implementation of Name() for enums that allow aliases in proto2 in Python to be in line with claims in C++ implementation (to return first value).
* Added double_format option in text format printer.
* Added iter and __contains__ to extension dict
* Added allow_unknown_field option in python text format parser
* Fixed Timestamp.ToDatetime() loses precision issue
* Support unknown field in text format printer.
* Float field will be convert to inf if bigger than struct.unpack('f', b'\xff\xff\x7f\x7f')[0] which is about 3.4028234664e+38,
convert to -inf if smaller than -3.4028234664e+38
* Allowed casting str->bytes in Message.__setstate__

Ruby
* Helper methods to get enum name for Ruby.

C++
* Use std::atomic<int32> in case of myriad2 platform
* Always declare enums to be int-sized
* Added DebugString() and ShortDebugString() methods on MessageLite
* Specialized different parse loop control flows
* Make hasbits potentially in register. The or's start forming an obstacle because it's a read modify store on the same mem address on each iteration.
* Move to an internal MACRO for parser validity checks.
* Improve map parsing performance.
* Make MergePartialFromCodedStream non virtual. This allows direct calls, potential inlining and is also a code health improvement
* Add an overall limit to parse_context to prevent reading past it. This allows to remove a annoying level of indirection.
* Fix a mistake, we shouldn't verify map key/value strings for utf8 in opt mode for proto2.
* Further improvements to cut binary size.
* Prepare to make MergePartialFromCodedStream non-virtual.
* A report on some interesting behavior change in python (caused by b/27494216) made me realize there is a check that needs to be done in case the parse ended on a end group tag.
* Add a note of caution to the comments around skip in CodedOutputStream.
* Simplify end check.
* Add overload for ParseMessage for MessageLite/Message types. If the explicit type is not known inlining won't help de-virtualizing the virtual call.
* Reduce linker input. It turns out that ParseMessage is not inlined, producing template instantiations that are used only once and save nothing but cost more.
* Improve the parser.
* [c++17] Changed proto2::RepeatedPtrField iterators to no longer derive from the deprecated std::iterator class.
* Change the default value of case_insensitive_enum_parsing to false for JsonStringToMessage.
* Add a warning if a field name doesn't match the style guide.
* Fix TextFormat not round-trip correctly when float value is max float.
* Added locationed info for some errors at compiler
* Python reserved keywords are now working with getattr()/setattr() for most descriptors.
* Added AllowUnknownField() in text_format
* Append '_' to C++ reserved keywords for message, enum, extension
* Fix MSVC warning C4244 in protobuf's parse_context.h.
* Updating Iterators to be compatible with C++17 in MSVC.
* Use capability annotation in mutex.h
* Fix "UndefinedBehaviorSanitizer: cfi-bad-type"
* CriticalSectionLock class as a lightweight replacement for std::mutex on Windows platforms.
* Removed vestigial wire_format_lite_inl.h

C
* Added System.Memory dependency.

Java
* Make Java protoc code generator ignore optimize_for LITE_RUNTIME. Users should instead use the Java lite protoc plugin.
* Change Extension getMessageDefaultInstance() to return Message instead of MessageLite.
* Prevent malicious input streams from leaking buffers for ByteString or ByteBuffer parsing.
* Release new Javalite runtime.
* Show warning in case potential file name conflict.
* Allow Java reserved keywords to be used in extensions.
* Added setAllowUnknownFields() in text format
* Add memoization to ExtensionRegistryLite.getEmptyRegistry()
* Improve performance of CodedOutputStream.writeUInt32NoTag
* Add an optimized mismatch-finding algorithm to UnsafeUtil.
* When serializing uint32 varints, check that we have MAX_VARINT32_SIZE bytes left, not just MAX_VARINT_SIZE.
* Minor optimization to RopeByteString.PieceIterator

JavaScript
* Simplify generated toObject code when the default value is used.

Python
* Changes implementation of Name() for enums that allow aliases in proto2 in Python to be in line with claims in C++ implementation (to return first value).
* Added double_format option in text format printer.
* Added iter and __contains__ to extension dict
* Added allow_unknown_field option in python text format parser
* Fixed Timestamp.ToDatetime() loses precision issue
* Support unknown field in text format printer.
* Float field will be convert to inf if bigger than struct.unpack('f', b'\xff\xff\x7f\x7f')[0] which is about 3.4028234664e+38,
convert to -inf if smaller than -3.4028234664e+38
* Allowed casting str->bytes in Message.__setstate__

Ruby
* Helper methods to get enum name for Ruby.

C++
* Avoid linking against libatomic in prebuilt protoc binaries (5875)
* Avoid marking generated C++ messages as final, though we will do this in a future release (5928)
* Miscellaneous build fixes

JavaScript
* Fixed redefinition of global variable f (5932)

Ruby
* Replace strptime with custom implementation (5906)
* Fixed the bug that bytes fields cannot be larger than 16000 bytes (5924)
* Miscellaneous bug fixes

PHP
* Replace strptime with custom implementation (5906)
* Fixed the bug that bytes fields cannot be larger than 16000 bytes (5924)

C++
* Introduced new MOMI (maybe-outside-memory-interval) parser.
* Add an option to json_util to parse enum as case-insensitive. In the future, enum parsing in json_util will become case-sensitive.
* Added conformance test for enum aliases
* Added support for --cpp_out=speed:...
* Added use of C++ override keyword where appropriate
* Many other cleanups and fixes.

Java
* Fix illegal reflective access warning in JDK 9+
* Add BOM

Python
* Added Python 3.7 compatibility.
* Modified ParseFromString to return bytes parsed .
* Introduce Proto C API.
* FindFileContainingSymbol in descriptor pool is now able to find field and enum values.
* reflection.MakeClass() and reflection.ParseMessage() are deprecated.
* Added DescriptorPool.FindMethodByName() method in pure python (c extension alreay has it)
* Flipped proto3 to preserve unknown fields by default.
* Added support for memoryview in python3 proto message parsing.
* Added MergeFrom for repeated scalar fields in c extension (pure python already has it)
* Surrogates are now rejected at setters in python3.
* Added public unknown field API.
* RecursionLimit is also set to max if allow_oversize_protos is enabled.
* Disallow duplicate scalars in proto3 text_format parse.
* Fix some segment faults for c extension map field.

PHP
* Most issues for json encoding/decoding in the c extension have been fixed. There are still some edge cases not fixed. For more details, check conformance/failure_list_php_c.txt.
* Supports php 7.3
* Added helper methods to convert between enum values and names.
* Allow setting/getting wrapper message fields using primitive values.
* Various bug fixes.

Ruby
* Ruby 2.6 support.
* Drops support for ruby < 2.3.
* Most issues for json encoding/decoding in the c extension have been fixed. There are still some edge cases not fixed. For more details, check conformance/failure_list_ruby.txt.
* Json parsing can specify an option to ignore unknown fields: msg.decode_json(data, {ignore_unknown_fields: true}).
* Added support for proto2 syntax (partially).
* Various bug fixes.

C
* More support for FieldMask include merge, intersect and more.
* Increasing the default recursion limit to 100.
* Support loading FileDescriptors dynamically.
* Provide access to comments from descriptors.
* Added Any.Is method.
* Compatible with C 6
* Added IComparable and comparison operators on Timestamp.

Objective-C
* Add ability to introspect list of enum values (4678)
* Copy the value when setting message/data fields (5215)
* Support suppressing the objc package prefix checks on a list of files (5309)
* More complete keyword and NSObject method (via categories) checks for field names, can result in more fields being rename, but avoids the collisions at runtime (5289)
* Small fixes to TextFormat generation for extensions (5362)
* Provide more details/context in deprecation messages (5412)
* Array/Dictionary enumeration blocks NS_NOESCAPE annotation for Swift (5421)
* Properly annotate extensions for ARC when their names imply behaviors (5427)
* Enum alias name collision improvements (5480)