John Keiser

Hmm, this really is a major usability issue. It's the sort of sharp edge you write disgruntled blog posts about after using simdjson for a few weeks or months. It's...

I'm all for processing the document in chunks, even in stage 1. It would make it more feasible to cut things into *more* stages that do less work, getting rid...

I don't think it would be much easier if it had the start and end of atoms in particular ... it would save a few instructions where we check for...

Same question as #670 : do we need input streaming for 1.0?

Moving out to 1.0 until we decide whether or not to block on it.

@lemire do you have any thoughts about how to do this? It seems like the biggest usability issue remaining in the library, sometimes genuinely affecting performance for users who get...

(I think it has become a barrier to adoption for some--it's definitely consistently misunderstood, and our most-asked question by far at the moment.)

Depending how it's implemented, too, it would let us safely expose a conversion from `document` to `value`, solving the annoying template problem.

Yeah, I think it's very reasonable to make it 1.0. It's clearly a barrier to adoption. That plan could work ... I think we are already using too much memory,...

I'm looking into what happens if we just special case the last 32 bytes in the frontend--i.e. how much of a hit we take adding this comparison.