v0.3.2

• Change signed varint encoding to signed-magnitude via varIntSignedPositive / varIntSignedNegative tags (replaces the need for ZigZag encoding).
• Add compact tags for very small values:
  • smallIntPositive / smallIntNegative (inline magnitude 0..7 in tag byte)
  • smallUint (inline unsigned 0..7 in tag byte)
  • smallBytes (inline length 0..7 in tag byte)
• Add typedArray for packed numeric vectors.
• Add native f16 float value encoding.
• Fix Reader.readPath / Reader.readPaths object size limit enforcement when encountering non-byte keys.
• Refactor code to reduce duplication.

v0.3.1

• Added Reader.readPath for fast path-based queries on encoded data using Javascript-style property access: "name", "address.city", "items[0]", "users[5].scores[0]".
• Added Reader.readPaths for efficiently reading a list of paths in a single pass.
• Added Writer.applyUpdates for efficiently applying a changeset of updates to an existing encoded buffer in a single pass.

Improvements

• Refactored and simplified some logic in Reader.zig

v0.3.0

• Upgrade to Zig 0.15.2
• Writer now accepts a *std.Io.Writer instead of an allocator.
• Inspect now accepts a *std.Io.Writer instead of the deprecated std.io.AnyWriter.
• Removed Writer.deinit(), Writer.bytes(), Writer.len(), Writer.toOwnedSlice() in favor of std.Io.Writer.

Safety

• Added configurable ReadLimits with the following options:
  • max_depth: maximum nesting depth for containers
  • max_bytes_length: maximum string/binary blob size
  • max_array_length: maximum array element count
  • max_object_size: maximum object key-value pair count
• New errors: MaxDepthExceeded, BytesTooLong, ArrayTooLarge, ObjectTooLarge.
• Inspect.printValue now returns NonFiniteFloat error for NaN and Infinity float value.

Benchmarks

• A benchmark suite has been added (mostly ported from zig-msgpack).

Bugfixes and Optimizations

• Fix integer overflow in Reader bounds checks that could bypass validation with malicious length values.
• Fix Reader depth underflow when encountering containerEnd at depth 0.
• Fix Reader float decoding to use little-endian byte order always.
• Fix Inspect JSON output to escape all control characters (0x00-0x1F).
• Fix encodeVarInt computing wrong size on big-endian machines.
• Inspect now returns error.InvalidUtf8 for non-UTF-8 byte sequences instead of producing invalid JSON.
• Fix integer overflow in Writer.write when writing extremely large integers.
• Fix Writer.writeAny silently dropping pointer-to-array values like &[_]u8{1,2,3} (data loss).
• Optimize: pre-compute key field for struct serialization.

v0.2.1

• Compatible with Zig 0.14.1
• Fix an issue in build.zig preventing compilation in macOS hosts.

v0.2.0

• Upgrade to Zig 0.14
• Simple variable length integer encoding (strictly better space efficiency) by prefixing length in the tag byte.
• Use ZigZag algorithm for efficiently encoding negative variable integers.
• Rename string type to bytes because a string is just an array of bytes. Also makes it clear that the type is meant for arbitrary binary values, not just strings. We don't want multiple "bytes" types like in MessagePack.
• bool is now stored directly inside it's tag, saving 1 byte per bool.
• Change runtime error.UnsupportedTypes to compile errors where possible.
• Rename to bufzilla because the old name was similar to Z-buffers (term in 3-D graphics programming).

v0.1.0

update readme