json_reader

Python

class JSONReader
JSONReader.formats() -> list[str]
JSONReader.extensions() -> list[str]
JSONReader.open(blob_path: Any) -> Any
JSONReader.read_native(handle: Any, variables: Sequence[str] | None = None, *, records_path: Sequence[str] | None = None, numeric_columns: Sequence[str] | None = None, **unknown: Any) -> NativeDataset

The active json reader: a JSON array of records becomes a points table.

The records are the array at records_path (a list of object keys from the document root; default [], the root itself). Each record must be an object. Its fields are its leaves: a nested object is flattened, its keys joined with . ({"coordinates": {"latitude": 1}} is the field coordinates.latitude). The fields are the union over all records; a field absent from a record is null there. Each field is one rank-1 array on the dimension index (length = number of records); there are no coordinates.

A field’s type comes from its non-null values: all numbers → float64 (null → NaN); all strings → string (null → ""); all booleans → bool (a null is an error, a boolean has no missing value); all null → float64 of NaN. A field mixing kinds is an error. A field named in numeric_columns is float64 whatever its cells: a number as is, null → NaN, a string holding a decimal number (after trimming ASCII whitespace) parsed, anything else an error. variables projects the fields; a field that is not decoded may hold anything, an array included.

Members

• read_native — Decode a JSON blob into a NativeDataset of points.

Source

Julia

struct JSONReader <: Reader
JSONReader()

JSONReader()

The json format reader: a JSON array of records becomes a points table, as the Python JSONReader and the Rust JsonReader decode it (one contract, pinned by conformance/corpus/scenarios/json-reader.json; spec/registries.md §2).

The records are the array at records_path (object keys from the document root; default none, the root itself). Each record must be an object; its fields are its leaves, a nested object flattened with its keys joined by . (coordinates.latitude). The fields are the union over the records; a field a record lacks is null there. Each field is one rank-1 array on index; there are no coordinates.

A field’s type comes from its non-null values: numbers → Float64 (null → NaN); strings → String (null → ""); booleans → Bool (a null is an error); all null → Float64 of NaN. Mixed kinds are an error. A field in numeric_columns is Float64 whatever its cells: a number as is, null → NaN, a string holding a decimal number (after trimming ASCII whitespace) parsed, anything else an error. variables projects the fields, and a field that is not decoded may hold anything, an array included.

Source

Rust

pub struct JsonReader {
    /* private fields */
}

impl JsonReader { pub fn new() -> Self }

impl JsonReader { pub fn records_path<I, S>(self, keys: I) -> Self where I: IntoIterator<Item = S>, S: Into<String> }

impl JsonReader { pub fn numeric_columns<I, S>(self, names: I) -> Self where I: IntoIterator<Item = S>, S: Into<String> }

The json format reader. The default instance reads the records from the document root and infers every field’s type.

Source · Rust docs