honey the codewitch

> This is nuance, but what happens if you try to iterate over an array expecting something like [ 1, 2, 3 ], but the JSON has a number (just...

On my tests from everything from 8 bit to 64 chunking didn't significantly impact performance, even with 5 byte buffers, but I imagine simd requires 128bit aligned data to be...

Okay I think I get that. You're just using branchlessness in every place you can to keep branch mispredictions down. And then you're using SIMD in places to cover some...

Because of complications dealing with JS escaped unicode surrogates, essentially, although after I wrote it I realized I could have just taken the 4 byte codepoint I got from the...

Regarding this: >Note: using simdjson's stage 1 might not give the full boost you'd expect. Much of stage 1's speed comes from avoiding branch misprediction when finding the values/structurals in...

I have a question of one of you, maybe you know. It's based around a hairbrained idea I had. Imagine you had a dummy func that just did a no-op...

I'm trying to make it so the function pointer that is put in the stream isn't placed there by an if. Basically what I intend is something like ``` *((pfnfetchdummy*!!(0!=mem)+(pfnfetchreal*!!(0==mem)))(pfetch_state);...

Wait, i think i get it. It's making a head hurt a little. :) so it would run along the branch where i called pfnfetchdummy before realizing it should have...

Okay I just figured it out I think. simdjson for bulk loading should forget streaming. What? Yeah. Forget streaming. Partition the file. Basically what you do is a stage 1...

I'm noticing simdjson requires a lot of upfront overhead. Sure it's impressive that it indexes a 20MB json file on my machine in about 46ms, but look at me searching...