// the find
miloyip/nativejson-benchmark
C/C++ JSON parser/generator benchmark
A benchmark suite that pits 40+ C/C++ JSON libraries against each other across parsing speed, memory usage, code size, and spec conformance. Built by the RapidJSON author, so take the framing with some salt, but the methodology is sound and the dataset coverage (real GeoJSON, Twitter data) is better than most synthetic benchmarks.
Measures conformance separately from performance — you see which libraries are fast but wrong, which matters if you're parsing untrusted input. Three realistic test files (canada.json with heavy floats, citm_catalog with deep nesting, twitter.json with CJK) expose pathological cases that contrived benchmarks miss. Tracks allocation count alongside time and peak memory, so you can identify libraries that thrash the allocator even when wall-clock numbers look fine. The HTML output with Google Charts lets you slice by benchmark type interactively rather than reading raw CSVs.
Last updated in 2016 for most libraries — simdjson, which rewrote the performance envelope entirely, appears as a submodule but the README doesn't mention it, so the headline numbers are badly stale. Build system requires premake5 plus manual submodule init across dozens of repos; getting a clean build in 2024 is genuinely painful and Travis CI is gone. The author's conflict of interest (RapidJSON is consistently shown at the top) is disclosed but not mitigated — there's no independent replication or cross-platform comparison baked in. No streaming/incremental parsing benchmark, which is the relevant case for large payloads in production.