explore [jsonlite](https://github.com/jeroenooms/JSONlite) for JSON translation

[timelyportfolio]

New package jsonlite or CRAN based off RJSONIO looks like a much better and speedier option for JSON translation.

[ramnathv]

I am finding it to be slower.

library(microbenchmark)

microbenchmark(
  jsonlite::toJSON(mtcars),
  RJSONIO::toJSON(mtcars),
  rjson::toJSON(mtcars),
  times = 10
)

Output

Unit: microseconds
                     expr        min         lq      median         uq        max neval
 jsonlite::toJSON(mtcars) 137884.341 142062.506 145352.0815 150982.395 177053.330    10
  RJSONIO::toJSON(mtcars)   1187.170   1280.926   1301.4975   1487.285   2240.849    10
    rjson::toJSON(mtcars)    226.272    230.432    271.4605    275.248    325.430    10

[timelyportfolio]

how odd. I was just downloading. I guess we should await a response on the jsonlite issue. Thanks for checking.

[ramnathv]

Yes. The toJSON function in jsonlite is more flexible and does both rows and columns. It also has a serializeJSON function which is super cool. I am surprised at the performance regression though since it is supposed to be based on RJSONIO.

[timelyportfolio]

I get a very simliar slow result for jsonlite on my Windows machine.

Unit: microseconds
                     expr        min         lq     median         uq      max neval
 jsonlite::toJSON(mtcars) 157140.033 160654.173 161913.867 164999.240 186773.6    10
  RJSONIO::toJSON(mtcars)   1536.035   1548.579   1681.807   1741.676 208468.5    10
    rjson::toJSON(mtcars)    568.648    620.344    633.458    687.244 325837.4    10

[timelyportfolio]

Agree jsonlite is a much more robust and helpful solution, but speed will be a big problem for now.

[jeroen]

Actually only the parser is based on RJSONIO. Both RJSONIO and jsonlite use plain R for toJSON.

[timelyportfolio]

jsonlite seems to be taking the spot as the standard for JSON import/export, and the newest version offers a big speed boost. Just a refresher to consider committing to jsonlite as the converter of choice.

[jeroen]

Just committed another performance tweak. Some benchmarks from the latest dev version:

> microbenchmark(
+   jsonlite::toJSON(diamonds),
+   jsonlite::toJSON(diamonds, dataframe="col"),
+   RJSONIO::toJSON(diamonds),
+   rjson::toJSON(diamonds),
+   times = 10
+ )

Unit: milliseconds
                                          expr      min       lq   median       uq      max neval
                    jsonlite::toJSON(diamonds) 761.1617 766.6734 776.6897 794.7294 810.3345    10
 jsonlite::toJSON(diamonds, dataframe = "col") 317.3538 323.0605 358.2399 369.7034 403.1690    10
                     RJSONIO::toJSON(diamonds) 868.9087 885.8630 893.2560 910.3101 952.7746    10
                       rjson::toJSON(diamonds) 306.6124 309.4605 309.8520 311.4325 324.0248    10

Note that rjson and RJSONIO by default use column based encoding for data frames. Only jsonlite supports row-based encoding. To benchmark performance make sure to compare to jsonlite::toJSON(x, dataframe="columns").

[ramnathv]

This is super-impressive @jeroenooms! I just tested it and the gains are very impressive. I am all for switching to jsonlite. The only thing keeping me from doing it is maintaining Shiny compatibility. I see from the Shiny issues thread that you are already discussing this.