John Keiser
John Keiser
I'm mentally moving towards either a one-pass compilation, or an n-pass compilation that compiles a c chunk at a time.
In general I just mean "a contiguous series of bytes from the input."
Chunk of text. Json is a hierarchical structure, but there's nothing saying you can't stop in the middle. I'm being vague on purpose because I don't really know the numbers....
It may be useful to limit short strings to 14 bytes or less, with no embedded nulls (not just terminating nulls), so that all "short strings" are guaranteed to be...
@mmatrosov that makes sense. I initially wanted to make sure dom::element and friends had to use move semantics, so that we could change the contract for efficiency without breaking people...
Yeah, this is a tough one in general. I think the removal of std::pair might have made it easier, but I think some seriously deep understanding of C++ is enough...
Fascinating. What kind of compression are we talking about? Not a subject I know well--I still sort of naively think of it as "identifying runs of n patterns and replacing...
So is the idea here that the buffer be compressed, and we decompress on demand (or streaming), getting indices from that (in stage 1), and then decompress on demand again...
There are some great ideas here. We've been discussing how to implement a true streaming interface to simdjson, and some of this dovetails nicely. > I think I can make...