Safe JSON serialisation

Safe serialisation of json with unboxing guided by the schema.

Usage

json_serialise(
  object,
  schema,
  engine = "ajv",
  reference = NULL,
  strict = FALSE
)

Arguments

object

An object to be serialised

schema

A schema (string or path to a string, suitable to be passed through to json_validator or a validator object itself.

engine

The engine to use. Only ajv is supported, and trying to use imjv will throw an error.

reference

Reference within schema to use for validating against a sub-schema instead of the full schema passed in. For example if the schema has a 'definitions' list including a definition for a 'Hello' object, one could pass "#/definitions/Hello" and the validator would check that the json is a valid "Hello" object. Only available if engine = "ajv".

strict

Set whether the schema should be parsed strictly or not. If in strict mode schemas will error to "prevent any unexpected behaviours or silently ignored mistakes in user schema". For example it will error if encounters unknown formats or unknown keywords. See https://ajv.js.org/strict-mode.html for details. Only available in engine = "ajv".

Value

A string, representing object in JSON format. As for jsonlite::toJSON we set the class attribute to be json to mark it as serialised json.

Details

When using jsonlite::toJSON we are forced to deal with the differences between R's types and those available in JSON. In particular:

• R has no scalar types so it is not clear if 1 should be serialised as a number or a vector of length 1; jsonlite provides support for "automatically unboxing" such values (assuming that length-1 vectors are scalars) or never unboxing them unless asked to using jsonlite::unbox

• JSON has no date/time values and there are many possible string representations.

• JSON has no data.frame or matrix type and there are several ways of representing these in JSON, all equally valid (e.g., row-wise, column-wise or as an array of objects).

• The handling of NULL and missing values (NA, NaN) are different

• We need to chose the number of digits to write numbers out at, balancing precision and storage.

These issues are somewhat lessened when we have a schema because we know what our target type looks like. This function attempts to use the schema to guide serialsation of json safely. Currently it only supports detecting the appropriate treatment of length-1 vectors, but we will expand functionality over time.

For a user, this function provides an argument-free replacement for jsonlite::toJSON, accepting an R object and returning a string with the JSON representation of the object. Internally the algorithm is:

1. serialise the object with jsonlite::toJSON, with auto_unbox = FALSE so that length-1 vectors are serialised as a length-1 arrays.

2. operating entirely within JavaScript, deserialise the object with JSON.parse, traverse the object and its schema simultaneously looking for length-1 arrays where the schema says there should be scalar value and unboxing these, and re-serialise with JSON.stringify

There are several limitations to our current approach, and not all unboxable values will be found - at the moment we know that schemas contained within a oneOf block (or similar) will not be recursed into.

Warning:

Direct use of this function will be slow! If you are going to serialise more than one or two objects with a single schema, you should use the serialise method of a json_schema object which you create once and pass around.

Examples

# This is the schema from ?json_validator
schema <- '{
    "$schema": "http://json-schema.org/draft-04/schema#",
    "title": "Product",
    "description": "A product from Acme\'s catalog",
    "type": "object",
    "properties": {
        "id": {
            "description": "The unique identifier for a product",
            "type": "integer"
        },
        "name": {
            "description": "Name of the product",
            "type": "string"
        },
        "price": {
            "type": "number",
            "minimum": 0,
            "exclusiveMinimum": true
        },
        "tags": {
            "type": "array",
            "items": {
                "type": "string"
            },
            "minItems": 1,
            "uniqueItems": true
        }
    },
    "required": ["id", "name", "price"]
}'

# We're going to use a validator object below
v <- jsonvalidate::json_validator(schema, "ajv")

# And this is some data that we might generate in R that we want to
# serialise using that schema
x <- list(id = 1, name = "apple", price = 0.50, tags = "fruit")

# If we serialise to json, then 'id', 'name' and "price' end up a
# length 1-arrays
jsonlite::toJSON(x)
#> {"id":[1],"name":["apple"],"price":[0.5],"tags":["fruit"]} 

# ...and that fails validation
v(jsonlite::toJSON(x))
#> [1] FALSE

# If we auto-unbox then 'fruit' ends up as a string and not an array,
# also failing validation:
jsonlite::toJSON(x, auto_unbox = TRUE)
#> {"id":1,"name":"apple","price":0.5,"tags":"fruit"} 
v(jsonlite::toJSON(x, auto_unbox = TRUE))
#> [1] FALSE

# Using json_serialise we can guide the serialisation process using
# the schema:
jsonvalidate::json_serialise(x, schema)
#> {"id":1,"name":"apple","price":0.5,"tags":["fruit"]} 

# ...and this way we do pass validation:
v(jsonvalidate::json_serialise(x, schema))
#> [1] TRUE

# It is typically much more efficient to construct a json_schema
# object first and do both operations with it:
obj <- jsonvalidate::json_schema$new(schema)
json <- obj$serialise(x)
obj$validate(json)
#> [1] TRUE