JSON custom numbers

https://github.com/jawj/json-custom-numbers

This package implements JSON parse and stringify functions to support custom number parsing and stringification.

Similar packages exist, but this one has some attractive features:

• Totally flexible number parsing and stringification, via functions you supply.

• Drop-in replacement for native JSON.parse and JSON.stringify: aims to exactly reproduce all other native behaviour.

• Non-recursive implementations, meaning deeply nested objects can't overflow the call stack.

• Faster than the alternatives.

• Informative error messages.

Note: the stringify() function makes use of native JSON.stringify() for string escaping, and is thus not a full replacement for the native function. If this is a problem for you, let me know: replace(), which is slightly slower, can easily be used for escaping instead.

Conformance and compatibility

The parse() function matches the behaviour of JSON.parse() for every test in the JSON Parsing Test Suite, and a few more besides.

The stringify() function matches JSON.stringify() for every valid case in the suite, and some others, with a variety of indent and replacer arguments.

Known differences

• The stringify() implementation is non-recursive, and the maximum allowable nesting depth is thus explicitly configurable (and defaults to 50,000). By contrast, native JSON.stringify() appears to be implemented with recursion, and maximum depth is thus limited by the call stack size.

• Error messages do not match any of the native implementations (which are all different).

If you discover any other behaviour that differs between these functions and the native JSON functions, please file an issue.

Performance

I've put some effort into optimising performance, and this library is substantially faster than similar libraries.

Performance comparisons depend both on the JavaScript engine and on the nature of the JSON data to be parsed/generated. If you figure out how to make either parse() or stringify() reliably faster, I'd be glad to hear about it.

Parse

In Node.js:

• In the best cases (all long strings with few escape sequences, or all deeply nested structures), this library may be slightly faster than JSON.parse().

• More usually, this library is 1.5 – 3x slower than JSON.parse(). This compares favourably with alternative packages, which can be up to 10x slower.

In Bun:

• This library is usually in the range 2 – 5x slower than JSON.parse(). This still compares favourably with alternative packages, which can be up to 20x slower.

• These numbers are worse than for Node.js both because Bun's native JSON.parse() is faster and because Bun runs this library slower.

Tests are included to compare the performance of this library, Crockford's reference implementation, and the json-bigint and lossless-json libraries against native JSON.parse across a range of inputs.

Reported timings represent a single parse() operation, and are the median of 50 trials of reps/50 operations each. Numbers in parentheses are the multiple of the time taken by native JSON.parse().

Lower numbers are better

Node.js 20.0.0 on a 2020 Intel MacBook Pro:

test               x   reps |   native |      this library |         crockford |       json-bigint |     lossless-json
01_typical_3kb     x  10000 |   11.4μs |   21.9μs  (x1.92) |   60.3μs  (x5.29) |   44.4μs  (x3.89) |   60.7μs  (x5.32)
02_typical_28kb    x   1000 |   99.0μs |  289.8μs  (x2.93) |  580.6μs  (x5.86) |  465.5μs  (x4.70) |  621.0μs  (x6.27)
03_mixed_83b       x  50000 |    1.8μs |    3.1μs  (x1.71) |    6.3μs  (x3.41) |    6.5μs  (x3.51) |    5.8μs  (x3.13)
04_short_numbers   x  50000 |    1.9μs |    6.2μs  (x3.32) |    8.6μs  (x4.63) |    8.7μs  (x4.66) |    7.9μs  (x4.26)
05_long_numbers    x  50000 |    1.8μs |    2.9μs  (x1.59) |    8.3μs  (x4.57) |   12.9μs  (x7.10) |    4.7μs  (x2.56)
06_short_strings   x  50000 |    1.9μs |    2.4μs  (x1.25) |    3.4μs  (x1.75) |    3.8μs  (x2.00) |    3.5μs  (x1.80)
07_long_strings    x   2500 |   54.7μs |   39.7μs  (x0.73) |  765.8μs (x14.00) |  529.6μs  (x9.68) |  455.9μs  (x8.33)
08_string_escapes  x 100000 |    1.0μs |    1.9μs  (x2.02) |   10.6μs (x11.02) |    9.9μs (x10.29) |    5.9μs  (x6.13)
09_bool_null       x 100000 |    0.9μs |    2.4μs  (x2.57) |    3.9μs  (x4.20) |    4.1μs  (x4.41) |    5.4μs  (x5.83)
10_package_json    x  25000 |    4.8μs |    8.0μs  (x1.68) |   36.2μs  (x7.56) |   29.5μs  (x6.16) |   25.4μs  (x5.31)
11_deep_nesting    x   1000 |  292.1μs |  280.7μs  (x0.96) |  533.7μs  (x1.83) |  504.6μs  (x1.73) |  606.0μs  (x2.07)
12_deep_indent     x   1000 |  362.1μs |  570.6μs  (x1.58) | 2249.7μs  (x6.21) | 2233.6μs  (x6.17) | 1667.0μs  (x4.60)

Bun 0.8.0 on a 2020 Intel MacBook Pro:

test               x   reps |   native |      this library |         crockford |       json-bigint |     lossless-json
01_typical_3kb     x  10000 |   12.9μs |   26.7μs  (x2.06) |   57.0μs  (x4.41) |   49.0μs  (x3.79) |   57.3μs  (x4.43)
02_typical_28kb    x   1000 |  101.5μs |  383.3μs  (x3.78) |  676.8μs  (x6.67) |  513.9μs  (x5.06) |  789.8μs  (x7.78)
03_mixed_83b       x  50000 |    1.7μs |    4.6μs  (x2.73) |    6.4μs  (x3.76) |    6.8μs  (x4.02) |    6.5μs  (x3.81)
04_short_numbers   x  50000 |    1.9μs |    8.7μs  (x4.52) |   12.3μs  (x6.42) |   11.9μs  (x6.18) |    7.9μs  (x4.10)
05_long_numbers    x  50000 |    1.1μs |    3.3μs  (x3.02) |   10.2μs  (x9.43) |   13.3μs (x12.27) |    4.6μs  (x4.24)
06_short_strings   x  50000 |    1.9μs |    5.7μs  (x3.07) |    4.3μs  (x2.28) |    4.4μs  (x2.33) |    3.9μs  (x2.11)
07_long_strings    x   2500 |   36.0μs |   47.6μs  (x1.32) | 1122.4μs (x31.17) |  437.9μs (x12.16) |  753.1μs (x20.91)
08_string_escapes  x 100000 |    1.0μs |    3.0μs  (x3.02) |    7.0μs  (x7.16) |    6.9μs  (x7.06) |    6.1μs  (x6.20)
09_bool_null       x 100000 |    0.8μs |    2.4μs  (x3.00) |    4.2μs  (x5.22) |    4.2μs  (x5.25) |    8.0μs (x10.00)
10_package_json    x  25000 |    4.8μs |    9.3μs  (x1.94) |   38.5μs  (x8.06) |   29.6μs  (x6.19) |   28.9μs  (x6.05)
11_deep_nesting    x   1000 |  152.2μs |  564.0μs  (x3.71) |  664.8μs  (x4.37) |  610.2μs  (x4.01) |  793.1μs  (x5.21)
12_deep_indent     x   1000 |  222.9μs | 1236.1μs  (x5.55) | 2101.9μs  (x9.43) | 1981.0μs  (x8.89) | 1714.7μs  (x7.69)

Stringify

The numbers for stringify() follow a more or less similar pattern, but performance differences between JSON.stringify(), this library and other libraries are generally smaller:

Lower numbers are better

Node.js 20.0.0 on a 2020 Intel MacBook Pro:

test               x   reps |   native |      this library |         crockford |       json-bigint |     lossless-json
01_typical_3kb     x  10000 |    8.2μs |   16.6μs  (x2.03) |   24.0μs  (x2.93) |   26.7μs  (x3.25) |   27.2μs  (x3.32)
02_typical_28kb    x   1000 |   59.0μs |  147.8μs  (x2.50) |  204.1μs  (x3.46) |  222.5μs  (x3.77) |  284.2μs  (x4.81)
03_mixed_83b       x  50000 |    1.5μs |    2.7μs  (x1.80) |    3.9μs  (x2.61) |    4.1μs  (x2.77) |    3.2μs  (x2.18)
04_short_numbers   x  50000 |    2.1μs |    3.6μs  (x1.73) |    4.5μs  (x2.16) |    5.6μs  (x2.68) |    7.1μs  (x3.39)
05_long_numbers    x  50000 |    2.0μs |    1.1μs  (x0.54) |    1.5μs  (x0.74) |    1.8μs  (x0.87) |    3.4μs  (x1.67)
06_short_strings   x  50000 |    1.1μs |    2.8μs  (x2.46) |    3.2μs  (x2.79) |    3.8μs  (x3.32) |    3.7μs  (x3.21)
07_long_strings    x   2500 |   97.9μs |  111.4μs  (x1.14) |   78.2μs  (x0.80) |   77.7μs  (x0.79) |  100.7μs  (x1.03)
08_string_escapes  x 100000 |    0.5μs |    0.6μs  (x1.19) |    3.4μs  (x6.30) |    3.4μs  (x6.33) |    0.6μs  (x1.04)
09_bool_null       x 100000 |    1.1μs |    1.6μs  (x1.48) |    2.5μs  (x2.33) |    2.9μs  (x2.72) |    4.2μs  (x3.85)
10_package_json    x  25000 |    4.4μs |    6.6μs  (x1.50) |    8.1μs  (x1.83) |    8.7μs  (x1.96) |   10.2μs  (x2.30)
11_deep_nesting    x   1000 |  155.6μs |  358.3μs  (x2.30) |  451.4μs  (x2.90) |  502.8μs  (x3.23) |  356.3μs  (x2.29)
12_deep_indent     x   1000 |  160.3μs |  364.8μs  (x2.28) |  455.8μs  (x2.84) |  503.7μs  (x3.14) |  357.6μs  (x2.23)

Bun 0.8.0 on a 2020 Intel MacBook Pro:

test               x   reps |   native |      this library |         crockford |       json-bigint |     lossless-json
01_typical_3kb     x  10000 |   10.5μs |   18.2μs  (x1.74) |   36.4μs  (x3.48) |   35.7μs  (x3.42) |   19.9μs  (x1.90)
02_typical_28kb    x   1000 |   93.6μs |  195.2μs  (x2.09) |  380.7μs  (x4.07) |  435.6μs  (x4.65) |  246.2μs  (x2.63)
03_mixed_83b       x  50000 |    1.8μs |    2.5μs  (x1.40) |    4.5μs  (x2.49) |    5.4μs  (x2.95) |    3.2μs  (x1.74)
04_short_numbers   x  50000 |    4.2μs |    4.0μs  (x0.94) |    5.5μs  (x1.31) |    7.0μs  (x1.65) |    8.0μs  (x1.89)
05_long_numbers    x  50000 |    1.2μs |    1.4μs  (x1.24) |    1.9μs  (x1.64) |    2.3μs  (x2.02) |    2.4μs  (x2.07)
06_short_strings   x  50000 |    0.5μs |    2.0μs  (x3.63) |    4.7μs  (x8.63) |    6.3μs (x11.74) |    2.6μs  (x4.89)
07_long_strings    x   2500 |   34.1μs |   45.9μs  (x1.34) |  108.3μs  (x3.17) |  107.3μs  (x3.14) |   32.8μs  (x0.96)
08_string_escapes  x 100000 |    0.6μs |    0.6μs  (x1.07) |    3.6μs  (x6.39) |    3.8μs  (x6.79) |    0.4μs  (x0.71)
09_bool_null       x 100000 |    0.5μs |    1.8μs  (x3.38) |    3.1μs  (x5.96) |    3.7μs  (x7.12) |    3.3μs  (x6.27)
10_package_json    x  25000 |    5.5μs |    5.3μs  (x0.96) |   12.3μs  (x2.23) |   14.8μs  (x2.69) |    6.7μs  (x1.22)
11_deep_nesting    x   1000 |  184.0μs |  352.8μs  (x1.92) |  768.6μs  (x4.18) |  955.2μs  (x5.19) |  326.6μs  (x1.78)
12_deep_indent     x   1000 |  174.1μs |  361.8μs  (x2.08) |  799.3μs  (x4.59) |  969.6μs  (x5.57) |  348.8μs  (x2.00)

Installation and use

Install:

npm install json-custom-numbers

Import:

import { parse, stringify } from 'json-custom-numbers';

For usage, see the examples below and the type definitions.

Parsing to BigInt

A key application of this library is converting large integers in JSON (e.g. from Postgres query results) to BigInts.

import { parse } from 'json-custom-numbers';

// `JSON.parse` loses precision for large integers
JSON.parse("9007199254740991"); // => 9007199254740991
JSON.parse("9007199254740993"); // => 9007199254740992 <- wrong number

// without a `numberParser` function, our behaviour is identical
parse("9007199254740991"); // => 9007199254740991
parse("9007199254740993"); // => 9007199254740992 <- wrong number

// this function converts only large integers to `BigInt`
function numberParser(k, s) {
  const n = +s;
  if (n >= Number.MIN_SAFE_INTEGER && n <= Number.MAX_SAFE_INTEGER) return n;
  if (s.indexOf('.') !== -1 || s.indexOf('e') !== -1 || s.indexOf('E') !== -1) return n;
  return BigInt(s);
}
parse("9007199254740991", null, numberParser);  // => 9007199254740991
parse("9007199254740993", null, numberParser);  // => 9007199254740993n <- now correct

Stringifying BigInt

In reverse:

import { stringify } from 'json-custom-numbers';

// this throws TypeError: Do not know how to serialize a BigInt
JSON.stringify(9007199254740993n);

// this serializes BigInt as a quoted string
JSON.stringify(9007199254740993n, (k, v) => typeof v === 'bigint' ? v.toString() : v);  // => "9007199254740993"

// this also serializes BigInt as a quoted string
BigInt.prototype.toJSON = function() { return this.toString(); }
JSON.stringify(9007199254740993n);  // => "9007199254740993"

// this serializes BigInt as a long number (i.e. unquoted), like Postgres does
function customSerializer(k, v, type) { if (type === 'bigint') return v.toString(); }
stringify(9007199254740993n, undefined, undefined, customSerializer);  // => 9007199254740993

Orientation

The code is in src/parse.ts and src/stringify.ts.

Currently, there are two build stages: the first creates .mjs files in src, while the second creates minified .js files in the root folder. The only package.json scripts you're likely to need to call directly are build and test/testConf/testPerf.

License

MIT licensed.

Note that most tests in the test_parsing folder that start with y_, n_ or i_ are from Nicolas Seriot's JSON Test Suite, which is also MIT licensed.