BSON Support

Glaze now supports BSON. Helper types (glz::bson::object_id, datetime, regex, javascript, decimal128, timestamp, binary, min_key, max_key) cover BSON wire types that have no native JSON equivalent, and glz::bson_to_json produces MongoDB Canonical Extended JSON v2.

• BSON read/write support by @stephenberry in #2518

Faster Float Writing

glz::to_chars for floats now uses a header-only port of Victor Zverovich's zmij (Schubfach + xjb) in place of dragonbox.

• ~2.7× faster than dragonbox on random finite doubles, ~3× faster on floats.
• The size-optimized path (OptSize=true) is ~4–8× faster than the previous simple_float::to_chars at roughly the same footprint.
• Side-effect bugfix: simple_float::from_chars no longer hits a 1-ULP rounding miss on platforms where long double aliases double (ARM64 macOS, MSVC).

Full exhaustive sweep over all 4,278,190,080 finite float values round-trips exactly.

• Replace dragonbox with zmij by @stephenberry in #2528

Struct Flattening with glz::merge

glz::merge now works inside glz::meta with member pointers, enabling both reading and writing of composite structs as flat JSON objects.

struct Species { std::string name{}; int legs{}; };
struct Appearance { double weight{}; std::string color{}; };
struct BearRecord { Species species{}; Appearance appearance{}; };

template <>
struct glz::meta<BearRecord> {
   using T = BearRecord;
   static constexpr auto value = glz::merge{&T::species, &T::appearance};
};

// Reads and writes as: {"name":"Grizzly","legs":4,"weight":300.5,"color":"brown"}

Sub-types can be either glaze_object_t (with custom key names) or reflectable (automatic reflection), and any number of sub-structs can be merged. Implemented entirely at the reflect<T> layer, so JSON, BEVE, YAML, CBOR, and other formats all pick it up automatically.

• glz::merge support for aggregate structs by @stephenberry in #2503

flatten_map Wrapper

A new glz::flatten_map wrapper serializes map-like containers as flat arrays of the form [key elements..., mapped value, ...], supporting fixed-size keys (std::pair, tuples, Glaze arrays, std::array) for both reading and writing.

std::unordered_map<std::pair<int, int>, std::string, pair_hash> map;
map[{1, 2}] = "example";

std::string out;
glz::write_json(glz::flatten_map_wrapper{map}, out);
// out == R"([1,2,"example"])"

Usable both inside glz::meta via member pointer (glz::flatten_map<&T::m>) and as a top-level CTAD wrapper (glz::flatten_map_wrapper{map}).

• flatten_map wrapper by @stephenberry in #2412
• flatten_map_wrapper top-level wrapping via CTAD by @stephenberry in #2526

Improvements

• Cross-format skip_null_members honored by JSONB, CBOR, MsgPack, and BEVE map writers, matching the existing struct-writer behavior by @stephenberry in #2518
• Remove force inlining for YAML and TOML to improve compile times and reduce binary size by @stephenberry in #2517

Fixes

• Named enum map keys: reading a std::map keyed by an enum declared with the keys + value array form of glz::meta no longer fails with expected_quote by @stephenberry in #2523
• jsonb_to_json now emits 9e999 / -9e999 for ±Infinity (matching SQLite's json() convention) instead of null. NaN continues to map to null. Documentation also clarifies that 64-bit integer / binary64 float ranges are a Glaze implementation constraint, not a SQLite spec limit by @stephenberry in #2521

Full Changelog: v7.4.0...v7.5.0