Docs/ Why Ion?

Dual-format interoperability

Applications can seamlessly consume Ion data in either its text or binary forms without loss of data fidelity. While the expectation is that most Ion data is in binary form, the text form promotes human readability, simplifying discovery and diagnosis.

Notably, the text format is a superset of JSON, making all JSON data valid Ion data. You probably already know how to read and author Ion.

Ease of use

Like JSON, Ion’s text format is concise and clearly legible. It is intelligible to humans and may be edited using familiar tools like a text editor. This makes Ion well-suited to rapid prototyping: developers can quickly mock up the their data in text Ion, knowing that their application will ultimately seamlessly process the more efficient binary Ion format. Once the application is up and running, the users can debug it by intercepting the binary Ion data and converting it to text Ion with full fidelity. After analysis, hand-edited records can be re-inserted into a processing pipeline as needed to support debugging and prototyping.

Text-only formats are more expensive to parse, which is why Ion offers the option of the length-prefixed binary format. This binary format supports rapid skip-scanning of data to materialize only key values within Ion streams.

Seamless transcoding

Ion’s interoperable formats avoid the kinds of semantic mismatches encountered when attempting to mix and match separate text and binary formats.

Standalone binary formats such as CBOR sacrifice human-readability in favor of an encoding that is more compact and efficient to parse. Although CBOR is based on JSON, transcoding between the two is not always straightforward because CBOR’s more expressive types do not necessarily map cleanly to JSON types. For example, CBOR’s bignum must be base-64 encoded and represented as a JSON string in order to avoid numeric overflow when read by a JSON parser, while a CBOR map may be directly converted to a JSON object only if all its keys are UTF-8 strings.

Rich type system

Ion’s type system is a superset of JSON’s: in addition to strings, Booleans, arrays (lists), objects (structs), and nulls, Ion adds support for arbitrary-precision timestamps, embedded binary values (blobs and clobs), and symbolic expressions. Ion also expands JSON’s number specification by defining distinct types for arbitrary-size integers, IEEE-754 binary floating point numbers, and infinite-precision decimals. Decimals are particularly useful for precision-sensitive applications such as financial transaction record-keeping. JSON’s number type is underspecified; in practice, many implementations represent all JSON numbers as fixed-precision base-2 floats, which are subject to rounding errors and other loss of precision.


Ion timestamps are W3C-compliant representations of calendar dates and time, supporting variable precision including year, month, day, hours, minutes, seconds, and fractional seconds. Ion timestamps may optionally encode a time zone offset.

By defining timestamps as a distinct type, Ion eliminated the ambiguity involved with representing dates as strings, as the semantics are clearly defined. Unlike a number, which counts from some “epoch”, arbitrary precision timestamps also allow applications to represent deliberate ambiguity.

Blobs and Clobs

Ion’s blob and clob types allow applications to tunnel binary data through Ion. This allows such applications to transmit opaque binary payloads (e.g. media, code, and non-UTF-8 text) in Ion without the need to apply additional processing to the payloads to make them conform to a different Ion type.

For example, a blob could be used to transmit a bitmap image, while a clob could be used to transmit Shift JIS text or an XML payload.

Symbolic expressions

The Ion specification defines a distinct syntax for symbolic expressions (S-expressions), but does not define how they should be processed. This allows applications to use S-expressions to convey domain-specific semantics in a first-class Ion type.

Formats that lack S-expressions as a first-class type are often left to choose between two imperfect options: adding a pre-processor (e.g. Jsonnet on top of JSON) to work around the inability to represent expressions as data, or tunneling domain-specific language text as opaque strings or binary payloads.


The Ion specification provides a formal mechanism for applications to annotate any Ion value without the need to enclose the value in a container. These annotations are not interpreted by Ion readers and may be used, for example, to add type information to a struct, time units to an integer or decimal value, or a description of the contents of a blob value.


Like JSON and CBOR, Ion is a self-describing format, meaning that it does not require external metadata (i.e. a schema) in order to interpret the structural characteristics of data denoted by the format. Notably, Ion payloads are free from build-time binding that inhibits independent innovation and evolution across service boundaries. This provides greater flexibility over schema-based formats such as protocol buffers, Thrift, and Avro, as data may be sparsely encoded and the implicit schema may be changed without explicit renegotiation of the schema among all consumers. These benefits come at the cost of a less compact encoding, but in our experience the positive impact on agility has been more valuable than an efficient but brittle contract.

Read-optimized binary format

Ion’s binary format is optimized according to the following principles:

In the spirit of these principles, the Ion specification includes features that make Ion’s binary encoding more efficient to read than other schema-free formats. These features include length-prefixing of binary values and Ion’s use of symbol tables.


Because most reads are sparse, binary Ion invests some encoding space to length-prefix each value in a stream. This makes seeking to the next relevant value for a particular application inexpensive, and enables efficient skip-scanning of data. This allows applications to cherry-pick only the relevant values from the stream for deeper parsing, and to economize parsing of irrelevant values.

Symbol tables

In binary Ion, common text tokens such as struct field names are automatically stored in a symbol table. This allows these tokens to be efficiently encoded as table offsets instead of repeated copies of the same text. As a further space optimization, symbol tables can be pre-shared between producer and consumer so that only the table name and version are included in the payload, eliminating the overhead involved with repeatedly defining the same symbols across multiple pieces of Ion data.

See also