Faster JSON parsing

[m-muecke]

Since the yyjsonr package is on CRAN now, I wanted to check the openess of having another option for parsing the JSON in req_body_json() with yyjsonr, since it seems to be quite a bit faster.

Quick benchmark:

str <- yyjsonr::write_json_str(iris)
bench::mark(
  jsonlite::fromJSON(str),
  yyjsonr::read_json_str(str)
)
#> # A tibble: 2 × 6
#>   expression                       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 jsonlite::fromJSON(str)      284.5µs  298.6µs     3298.    1.29MB    16.5 
#> 2 yyjsonr::read_json_str(str)   26.9µs   27.9µs    34519.    9.69KB     3.45

^{Created on 2024-11-14 with reprex v2.1.1}

[hadley]

I'm slightly worried that this might introduce subtle parsing changes, but it's probably going to be ok. I'll just need to think about how to parameterise and what should be the default.

[hadley]

• jsonlite is already in suggests, so adding yyjsonr there too would be fine.
• It looks like the parsing is different enough that I don't think we can just swap (i.e. yyjsonr doesn't seem to have an equivalent of auto_unbox.)
• If we do this all in one function, we'd probably need to add a parser argument and an options argument then deprecate auto_unbox, digits, and null and then create some helper that's the equivalent of opts_read_json.
• Does that suggest it's better to use a suffix? resp_body_json_jsonlite() vs resp_body_json_yyjsonr(). Those are long function names.
• Would be nice to use either read_json_raw() or read_json_file().

[m-muecke]

There is an auto_unbox equivalent in yyjsonr, in case you mean the following:

x <- list(x = 1, y = 1:10)
yyjsonr::write_json_str(x, opts = yyjsonr::opts_write_json(auto_unbox = TRUE))
#> [1] "{\"x\":1.0,\"y\":[1,2,3,4,5,6,7,8,9,10]}"

^{Created on 2024-12-20 with reprex v2.1.1}

Integers would have to be explicitly set or set via the digits argument.

[hadley]

@m-muecke that's for writing, not reading though.

[hadley]

Oh which is what we are doing here, duh.

[hadley]

I was thinking about resp_body_json(). Which do also need to think about.

[hadley]

All in all, I think given the difficulties of making this change on global basis, compared to using yyjsonr for your specific API, I don't think it's worth changing httr2 at this time. But I appreciate you bringing this up and thanks for using httr2!