A JSON parsing library for the Noir Language. This library adheres to the IETF RFC 8259 specifications (mostly, see edge_cases).
Features:
- handles arbitrary-length JSON strings (up to a defined maximum bound)
- flexible interface can process arbitrary JSON schemas
- O(1) algorithms to query whether keys exist
requires Noir v0.32 or greater
if using bb backend, megaplonk (
bb prove_mega_honk
) will give best performance as json-parser uses a megaplonk-friendly hash function Poseidon2 (see bb documentation for install/upgrade details)
Multiple JSON parser types are exposed that accomodate different maximum length parameters:
Notation explainer: a token is a distinct JSON element, either
{
}
[
]
,
:
, strings, numbers of literals (false
,true
,null
) Notation explainer: a value is an object, array, string, number of literal
JSON512b::JSON
a parser that can handle up to 512 bytes of JSON, up to 64 tokens and 32 valuesJSON1kb::JSON
a parser that can handle up to 1kb of JSON, up to 128 tokens and 64 valuesJSON2kb::JSON
a parser that can handle up to 2kb bytes of JSON, up to 256 tokens and 128 valuesJSON4kb::JSON
a parser that can handle up to 4kb bytes of JSON, up to 512 tokens and 256 valuesJSON8kb::JSON
a parser that can handle up to 8kb bytes of JSON, up to 1,024 tokens and 512 valuesJSON16kb::JSON
a parser that can handle up to 16kb bytes of JSON, up to 2,048 tokens and 1,024 values
The maximum length of keys in the JSON blob is 62 bytes. All of these parameters are configurable, see Advanced usage for more info.
use crate::json_parser::1kbJSON::JSON;
use crate::json_parser::JSONLiteral;
/*
example schema. In this example we require every field to exist except "dlc_enabled", which is optional.
We also demonstrate escaped strings. To escape a string in JSON, we must escape the escape characters in noir!
{
"hero": "Tony Harrison",
"is_player": false,
"stats": [
{
"hero_quote": "\\\"This is an outrage!\\\"",
"power_level": 1
},
{
"hero_quote": "\"The Crunch? You know nothing of the crunch!\"",
"power_level": 9001
}
],
"dlc_enabled": true
}
*/
fn process_schema(text: [u8; NUM_BYTES])
{
let json: JSON = JSON::parse_json(text);
// we use the "unchecked" method because we want to assert this field exists
let hero_name: BoundedVec<u8, 20> = json.get_string_unchecked("hero".as_array());
assert(hero_name == BoundedVec::from_array("Tony Harrison".to_array()));
// json literals can be "null", which doesn't map perfectly to a bool, so we use a custom type
let is_player: JSONLiteral = json.get_literal_unchecked("is_player".as_array());
let is_player = is_player.to_bool(); // can cast to bool if you don't mind null mapping to false
// to move into a nested object or array, call `get_object` or `get_array`
let stats_array: JSON = json.get_array_unchecked("stats".as_array());
let stats: [JSON; 2] = [stats.get_object_as_array_unchecked(0), stats.get_object_as_array_unchecked(1)];
let hero_quote: BoundedVec<u8, 30> = stats[0].get_string_unchecked("hero_quote".as_array());
assert(hero_quote == BoundedVec::from_array("\"This is an outrage!\"".as_array()));
let power_levels: [u64; 2] = [
stats[0].get_number_unchecked("power_level"),
stats[1].get_number_unchecked("power_level"),
];
assert(power_levels[1] > power_levels[0]);
// All getter methods have a basic version that returns the element as an Option,
// to support cases where a field may or may not exist
let dlc_enabled: Option<JSONLiteral> = json.get_literal("dlc_enabled");
let has_dlc_field = dlc_enabled.is_some();
}
- A single value is a valid JSON schema but is not currently supported.
i.e. let json: JSON = JSON::parse_json("9999")
will fail
Numbers must currently fit within a u64 for get_number
and associated methods to work.
In addition, numbers with decimal points currently are not supported, as well as numbers that use scientific notation (e.g. a json blob with 3e5
will create a failing proof)
For numbers larger than 64 bits, the method get_value
will work, which will return the ascii raw bytes of the value.
How much does this thing cost to use? Details below:
There is an up-front processing cost, as well as a per-query cost.
json type | cost to parse (in megaplonk gates) |
---|---|
512b | 42k |
1kb | 71k |
2kb | 128k |
4kb | 160k |
8kb | 307k |
16kb | 574k |
query type | query cost (in megaplonk gates) |
---|---|
get_value |
364 |
get_number |
454 |
get_literal |
416 |
get_string |
435 |
get_object |
165 |
get_array |
165 |
get_value_unchecked |
327 |
get_number_unchecked |
415 |
get_literal_unchecked |
346 |
get_string_unchecked |
547 |
get_object_unchecked |
143 |
get_array_unchecked |
143 |
If the predefined JSON types are not sufficient for your use case, you can define a JSON object with a custom parametrisation:
The JSON struct in dep::json_parser::json::JSON
is parametrised with the following parameters:
struct JSON<let NumBytes: u32, let NumPackedFields: u32, let MaxNumTokens: u32, let MaxNumValues: u32, let MaxKeyFields: u32>
NumBytes
: the maximum size of the initial json blobNumPackedFields
: takeNumBytes / 31
, round up to the nearest integer, then add 3!MaxNumTokens
: the maximum number of tokens in the json blobMaxNumValues
: the maximum number of values in the json blob plus oneMaxKeyFields
: the largest size a key can take (in bytes) equalsMaxKeyFields * 31
e.g. to take the existing 1kb JSON paramters, but also support 124-byte keys, use JSON<1024, 37, 128, 65, 4>
If you are deriving a key to look up in-circuit and you do not know the maximum length of the key, all query methods have a version with a _var
suffix (e.g. JSON::get_string_var
), which accepts the key as a BoundedVec
Many thanks to the authors of the OG noir json library https://github.com/RontoSOFT/noir-json-parser