Daniel Lemire

It looks fine at a glance. We do support `string_view` throughout in simdjson and it is a standard approach. To my knowledge, we have no rule against using STL. We...

The `emplace` for each number is likely to come at a significant burden in some cases. It would be safer to make this mechanism a compile-time option though it will...

I think that there might be a much more performant approach that would be effectively free. https://github.com/simdjson/simdjson/blob/master/doc/tape.md#integer-and-double-values You have 56 unused bits per number, and we need only 32 bits...

@michaeleisel We hand out an in-document to the user and the user is responsible for checking in the document source. So the library has minimal responsibility. Best not to hand...

The starting location of the number string is probably sufficient.

@michaeleisel I expect a small increase in instruction count and possibly a small decrease in performance, but it should be effectively hard to notice. Before merging, I will review that...

@michaeleisel Please review the CI test results.

@michaeleisel Nitpicking. Is `get_json_index_of_number` the best we can come up with? It does not seem like an intuitive name to me. - `get_location_of_number_in_json` - `get_byte_index_of_number_in_json` I hate bike shedding so...

@michaeleisel I expect to merge this. It is well done all around. There will be a small price to pay but it should be fine. The regression is already measurable,...

@michaeleisel Would you sync with our main branch ? It appears that one of the test plaform has overeager static checks, and it explains why you have a failing test....