v2.6.5

Fixes

• Fixed floating point writing for std::vector<std::byte>. Now guarding against implicit conversions internally.
  by @stephenberry in #1009 & #1011

Full Changelog: v2.6.4...v2.6.5

v2.6.4

Fixed support for Clang 16, 17, & 18 (v2.6.3 fix for GCC 14 had broken Clang)

• Glaze now has actions to build and test on Clang 16, 17, and 18 on Linux, so these issues should be avoided in the future

by @stephenberry in #1006

Improvements

• Improved write performance when objects write out all fields
• Improved write performance for some types of known sizes

Fixed

• Removed undefined behavior when using memcpy on a null pointer over zero bytes

Full Changelog: v2.6.3...v2.6.4

v2.6.3

Support for GCC 14

• GCC 11 maintenance is being dropped. Glaze will continue to ensure compatibility with GCC 12, 13, and 14.

by @stephenberry in #1001

Fixes

• Fixed a potential buffer under-read when using const read buffers that are extremely short (less than 12 characters).

Improvements

• More larger data set unit tests

Full Changelog: v2.6.2...v2.6.3

v2.6.2

Improvements

• CRLF (carriage return) support in CSV by @stephenberry in #980
• Faster string read/write by @stephenberry in #991
• Removed read/write file inconsistencies by @stephenberry in #992
• Moving raw_buffer and output_buffer to container_concepts by @stephenberry in #993

Fixes

• Fix asio socket->set_option by @stephenberry in #994

Thanks to everyone who contributed feedback!

Full Changelog: v2.6.1...v2.6.2

2.6.1

Important

Fixed UTF-8 reading bug that affected validation and proper parsing. (by @stephenberry in #979)

Other Fixes

• Fix for glz::skip and compile time errors when invalid by @stephenberry in #970

Improvements

• Faster string assignment (avoid null memset on resize) by @stephenberry in #968

Full Changelog: v2.6.0...v2.6.1

2.6.0

Major Update

Potentially breaking binary (BEVE) read changes

The binary error checking is becoming much more robust, with the intent to make it completely safe for untrusted inputs. Currently many, but not all, input errors are handled, so binary should only be used in trusted contexts, where the developer is handling both encoding and decoding.

Important

This update requires BEVE input buffers to be null terminated. Just like Glaze requires null termination for JSON input buffers.

• Allows fast validation
• Allows common buffer handling between binary and JSON

If you were using std::vector<std::byte> for your binary input buffer, simply append a null character to your buffer before reading.

It is recommended to use a std::string for binary reading, as std::string ensure null termination.

Padding for Mutable Input Buffers

When non-const (mutable) input buffers are passed in (e.g. std::string versus const std::string), Glaze now temporarily adds padding to improve SIMD performance and reduce assembly.

This is not a breaking change. It is simply recommended to pass mutable input buffers into read calls.

The buffer is padded with null bytes while reading and then reverted at the end of the function call. Glaze does not permanently change your input buffer.

• This change reduces the produced assembly, so the overall program binary will be reduced when using mutable input buffers.

Improvements

• Faster glz::minify_json
• Faster string reading for escaped inputs
• Faster unknown key discover in some cases
• More specific error handling for commas, colons, and quotes

minfied Compile Time Option

If you wish to require minified JSON or know your input will always be minified, then you can gain a little more performance by using the compile time option .minified = true.

auto ec = glz::read<glz::opts{.minified = true}>(obj, buffer);

Fixes

• Proper tab usage in glz::prettify_json
• Using full 256 byte tables for safely handling invalid JSON

Removed Deprecations

• Removed the deprecated macros.hpp header
• Removed the deprecated glz::prettify and glz::minify functions. Use the new glz::prettify_json and glz::minify_json

Full Changelog: v2.5.5...v2.6.0

2.5.5

Improvements

• A recent change made vector reading slower for small types in std::vector, the higher performance for small types has been restored with the improvements to large types (greater than 4096 bytes) intact
• Adding more binary checks and notes on binary safety
• Code cleanup

Full Changelog: v2.5.4...v2.5.5

2.5.4

Support unicode surrogate pairs

As per the latest JSON standard, unicode pairs can be used to encode UTF-16 characters in UTF-8 JSON. This release adds support for surrogate pairs.

const char* json = R"("\uD83D\uDE00\uD83C\uDF40😀🍀\uD83D\uDE00")"; // 😀🍀😀🍀😀
std::string val;
expect(!glz::read_json(val, json));
expect(val == "😀🍀😀🍀😀");

Float max write precision

Adds the ability to set the maximum write precision for floating point values. Wrappers are also provided to locally turn this feature on and off. This reduces the size of the JSON output for floating point numbers when the precision isn't needed on output and is especially useful for std::float128_t. See documentation at max-float-precision.md

Examples:

double pi = std::numbers::pi_v<double>;
std::string json_double = glz::write_json(pi);

constexpr glz::opts options{.float_max_write_precision = glz::float_precision::float32};
std::string json_float = glz::write<options>(pi);
expect(json_float == glz::write_json(std::numbers::pi_v<float>));

std::vector<double> double_array{pi, 2 * pi};
json_double = glz::write_json(double_array);
json_float = glz::write<options>(double_array);
expect(json_float == glz::write_json(std::array{std::numbers::pi_v<float>, 2 * std::numbers::pi_v<float>}));

Improvements

• Using 256 byte tables for invalid JSON safety by @stephenberry in #937
• Faster escaped unicode handling

Full Changelog: v2.5.3...v2.5.4

2.5.3

Improvements

• Faster string comparison in #925
• Cleaned/optimized error message generation and includes in #927
  • Uses less allocations/memory as well, which is good for embedded platforms
• Support tag property in tagged variants by in #929
• Better errors when attempting to write to a const buffer in #933

Internal Cleanup

• Removed unused tuplet _tag in #924

Full Changelog: v2.5.2...v2.5.3

2.5.2

Fixes

• glz::get_as_json fixed for intermediate array access by @stephenberry in #923
  • Much improved error handling for the function

Improvements

• Faster skipping
• Faster whitespace parsing in common contexts

Full Changelog: v2.5.1...v2.5.2