zugojson

zugojson formats, minifies and validates JSON text. Its parser and printer are written in Goose, a memory-safe systems language with no heap, and compiled ahead of time to C. The package ships that C, so installing it needs no Goose compiler.

The package is a demonstration that Goose can be used to write R packages. It is not a replacement for jsonlite.

Installation

# install.packages("pak")
pak::pak("pedrobtz/zugojson")

zugojson builds on Linux and macOS. Windows is not supported yet.

Usage

library(zugojson)

cat(json_format('{"a": [1, 2.50, {"b": null}], "s": "x"}'))
#> {
#>   "a": [
#>     1,
#>     2.5,
#>     {
#>       "b": null
#>     }
#>   ],
#>   "s": "x"
#> }

json_minify('{ "a" : [ 1, 2 ] }')
#> [1] "{\"a\":[1,2]}"

json_valid(c("[1, 2]", "[1, 2", NA))
#> [1]  TRUE FALSE    NA

A syntax error is a condition of class zugojson_parse_error. Its byte field holds the offset where parsing stopped.

tryCatch(json_format("[1, 2"), zugojson_parse_error = function(e) e$byte)
#> [1] 5

How it works

The JSON code is in tools/goose/zugojson.goose. tools/regenerate.sh compiles it with a pinned Goose compiler into src/zugojson_goose.c, a C file with no main.

The parsed document lives in one contiguous block. Each node links to its next sibling with a 4-byte offset relative to itself, so the tree holds no pointers.

Generated Goose code needs a runtime. The stock runtime ends the process on any error, writes to stdout and stderr, and installs process-wide signal handlers. That is unacceptable inside R, so zugojson replaces it with src/goose_r_runtime.c:

• A Goose runtime error, such as a failed bounds check, becomes an R error of class zugojson_runtime_error, and later calls keep working.
• Goose's guard-page fault handler is installed only while Goose code runs. R's own handlers are restored afterwards.
• Nothing writes to stdout or stderr, and nothing can end the R process.

Performance

zugojson is about as fast as jsonlite 2.0.0. The table gives seconds for four documents of 1 to 4 MB, built with jsonlite's toJSON(): an array of integers, one of random doubles, one of short strings and a table of records. Each time is the best of five runs, with R 4.5.2 on macOS.

Document	Task	zugojson	jsonlite
Integers	Minify	0.017 s	0.014 s
	Pretty-print	0.019 s	0.020 s
	Validate	0.013 s	0.009 s
Doubles	Minify	0.033 s	0.028 s
	Pretty-print	0.035 s	0.034 s
	Validate	0.019 s	0.018 s
Strings	Minify	0.007 s	0.008 s
	Pretty-print	0.007 s	0.010 s
	Validate	0.003 s	0.004 s
Records	Minify	0.020 s	0.032 s
	Pretty-print	0.023 s	0.044 s
	Validate	0.012 s	0.016 s

Licence

zugojson is MIT licensed. The Goose-derived files are under Apache-2.0, as listed in inst/COPYRIGHTS.