v1.2.0

Shared dasmig::sex (core/sex.hpp: sex_name, parse_sex) and country code helpers (core/country.hpp: is_alpha2, unknown_country_error). Additive; generated output is unchanged.

v1.1.1

Compact key index and TSV cells, and a sized file read: name-generator's full tier loads in 1.1 s instead of 1.7 s and holds 267 MiB instead of 593 MiB. Output is unchanged.

v1.1.0

What's Changed

• feat: batch generation, JSON Lines and CSV writers (1.1.0) by @dasmig in #4

Full Changelog: v1.0.2...v1.1.0

v1.0.2

What's Changed

• docs: put the MIT notice in every header by @dasmig in #2
• perf: fewer allocations in json_writer and weighted_table (1.0.2) by @dasmig in #3

Full Changelog: v1.0.1...v1.0.2

v1.0.1

Faster TSV parsing, identical output. Libraries keep DASMIG_CORE_REQUIRE(1, 0); pick it up with tools/sync-core.sh.

Changed

• tsv parses plain decimals through an exact fast path (Clinger's): when the significant digits form an integer up to 2^53 and the power of ten is within ±22, one IEEE multiply or divide gives the correctly rounded double. Anything else, and platforms with excess precision (FLT_EVAL_METHOD != 0), use the general parser as before.
• Why: Emscripten's libc++ has no floating-point from_chars, so every cell went through istringstream and musl's software long double strtod. That was about 45% of the persona WebAssembly module's startup, which drops from 1.4 s to 0.35 s in Node 22.

Output is identical

• A new test compares the parser with strtod bit for bit on edge cases and 50,000 random decimals.
• All 4.94 million numeric cells in the nine libraries' data (both tiers) parse to the same bits; 99.99% take the fast path.
• The persona module built with this core gives the same 850 personas and data hash.

v1.0.0

First stable release. The seven dasmig generator libraries now build on the core, so its output contract is fixed for major version 1: a seed gives the same bits on every supported compiler, standard library and CPU (GCC, Clang with libc++, Apple Clang, MSVC, Emscripten; x86-64 and ARM64), checked by known-answer tests in CI.

Output is identical to 0.3.4. Libraries require it with DASMIG_CORE_REQUIRE(1, 0).

Contents

• rng: xoshiro256** seeded through SplitMix64, with split(label) / split(index) for independent streams.
• Distributions with fully specified algorithms: uniform_int (Lemire), uniform_real, bernoulli, normal (Marsaglia polar), lognormal, discrete, shuffle, sample_indices, pick; own exp/log and FMA-proof arithmetic.
• tsv reader with typed columns and file:line: column errors; layered_table for data layers (add / override / replace, disabled rows, all-or-nothing validation, data_hash()).
• weighted_table, json_writer, seed_signature, utf8, dataset, resources (resource probing and compiled-in data; scripts/embed.py).

Since 0.3.3

• layered_table indexes keys once, in a hash map: loading name-generator's 2.3 million rows went from 2.5 s to 0.67 s in the table, with the same rows, order, hashes and errors (0.3.4).