0.5.0

Breaking changes

As many people have been using D:YAML from git master since the 0.4 release, each change is prefixed by the year the change was introduced.

2014 Broke compatibility with all DMD versions before 2.066
2014 Loader API depending on std.stream is now deprecated and will be

removed in the next release.
2014 Loader.fromString(string) is now deprecated and replaced by

Loader.fromString(char[]), which will reuse and overwrite the input during parsing. The string overload will be removed in the next release.
2012 Values in D:YAML are less 'dynamic-typed'. E.g. removing "42" won't

also remove 42. Such automatic conversion still happens e.g. in a foreach over a mapping with a string key. The Node.as() method has a template parameter to disable automatic conversion from strings to other types.

Loader is now, by default, constructed from a non-const buffer containing YAML data instead of a std.stream.Stream. D:YAML reuses and overwrites this buffer to minimize memory allocations during parsing. Convenience API such as the Loader constructor from a filename is unchanged although it loads the file to a buffer internally.
Node.contains() method to check if a YAML sequence/mapping contains specified value (as opposed to a key).
Node.containsKey() method to check if a YAML mapping contains specified key.
Node.isNull() property.
Node operator in analogous to associative array in.
Loader.fromString() method as a quick wrapper to parse YAML from a string.
dyaml.hacks module for potentially useful things YAML specification doesn't allow.
Most of the API is @safe or at least @trusted. Also pure and nothrow where possible.
User-defined constructors can now also construct builtin YAML types.
Input is now validated at Loader construction to detect invalid UTF sequences early and to minimize internal exception handling during parsing.

D:YAML now works with a UTF-8 buffer internally. This decreases memory usage for UTF-8 input, and UTF-32 inputs can be encoded into UTF-8 in-place without allocating. UTF-16 inputs still need an allocation. This also gets rid of all dchar[]->char[] conversions which were a significant source of GC allocations.
Less GC usage during YAML node construction.
The Scanner subsystems is now mostly @nogc; it never allocates memory for token values, using slices into the input buffer instead.
Custom, @nogc UTF decoding/encoding code based on std.utf to enable more use of @nogc in D:YAML internals and to improve performance.
Replaced std.stream.EndianStream with custom endian-fixing code.
D:YAML is now a DUB package.
Removed static data structures such as default Constructor and Resolver.
Compile-time checks for size of data structures that should be small.
Better error messages.
Various refactoring changes, using more 'modern' D features, better tests.
Updated documentation, examples.

Fixed mappings longer than 65535 lines.
Removed a lot of in parameters that were used due to a misunderstanding of what in is supposed to do.
Fixed real emitting.
Fixed 32bit compilation (again).
Various small bugfixes.

API BREAKING: All structs and classes stored directly in YAML nodes (aka custom YAML data types) now need to define the opCmp operator. This is used for duplicate detection instead of AAs (which are broken) and will allow efficient access to data in unordered maps.
API BREAKING: Simplified the Constructor API. Constructor functions now don't need to get Marks through parameters - any exceptions thrown by the constructor functions are caught and wrapped along with Mark info.
Various small improvements in the API documentation.
Updated API documentation, tutorials and examples to match the changes.
Small CDC (build script) improvements.

API BREAKING: Removed Node.getToVar as it turned out to be a premature optimization.
API BREAKING: Constructor API for constructing custom YAML data types has been improved to make it easier to load custom classes/structs. See the custom types tutorial and Constructor API documentation.
Node.opIndex now returns a reference to a node.
Added a shortcut alias Node.as for Node.get . Node.as might eventually replace Node.get (in a 1.0 release).
User can now access a string representation of tag of a node.
Nodes now remember their scalar and collection styles between loading and dumping. These are not accessible to user. User can set output styles in Representer.
Updated API documentation to reflect the new changes, added more examples and overall made the documentation more readable.
Improved error messages of exceptions.
Drastically optimized scanning and parsing, decreasing parsing time to about 10% (no precise benchmark comparison with 0.2 at the moment).
Eliminated most GC usage, improving speed and memory usage.
Optimized Dumper for speed, especially when dumping many small files.
Reader has been reimplemented to improve performance.
Many other speed and memory optimizations.
Added a profiling build target and a parsing/dumping benchmark.
Added a random YAML file generator and a YAML file analyzer, as example applications and for benchmarking.
Added a "clean" target to example Makefiles.
Got rid of all global state.

Fixed compatibility issues with DMD 2.056.
Fixed an Emitter bug which caused tags to always be emitted in full format.
Fixed a bug that caused errors when loading documents with YAML version directives.
Fixed many const-correctness bugs.
Minor bugfixes all over the code.
Fixed many documentation bugs.

Implemented YAML emitter, and related unittests/documentation.
Tags are now stored in nodes, allowing D:YAML to be closer to the specification.
Loader API has been broken to make it more extensible in future - Representer and Constructor are no more specified in the constructor, and the load() shortcut functions have been removed, as all that's needed to load a YAML document now is Loader("file.yaml").load() .

Fixed many bugs in the parser, scanner, composer and constructor.