Timing test shows yyjson vs glaze at 14ms vs 700ms

[Please-just-dont]

I tried parsing this file with with the yyjson C library. I did yyjson_read vs glz::read_json and I got 14 milliseconds for yyjson and 700 milliseconds for glaze. Does that sound right?

Edit: OK, so I thought maybe glaze is actually parsing the json into its own structures whereas yyjson is just doing like a prepass or something, so I tested by reading into my own structures with glz::json_t* and ->get functions, and the results turned out:

yyjson: 3.5ms
glaze 15ms

So still. In favour of yyjson by a big margin.

[stephenberry]

glz::json_t is not going to get you peak performance in Glaze. glz::json_t is only for structures that you know nothing about and therefore requires dynamic memory allocations. Your yyjson code can likely not accept any JSON like glz::json_t.

You should put your data into C++ structs, this makes the keys known at compile time, and reduces dynamic allocations to dynamic arrays. Glaze will likely outperform yyjson in this scenario. Though, there are additional considerations for peak performance, such as using a mutable std::string for the input buffer so that Glaze can pad the input and do more efficient SIMD.