JavaScript Object Notation (JSON) Text Sequences
A way to stream many JSON values one after another, with a marker byte that lets readers recover from damage.
Williams, N., "JavaScript Object Notation (JSON) Text Sequences", RFC 7464, DOI 10.17487/RFC7464, February 2015, <https://www.rfc-editor.org/info/rfc7464>.
@misc{rfc7464,
series = {Request for Comments},
number = 7464,
howpublished = {RFC 7464},
publisher = {RFC Editor},
doi = {10.17487/RFC7464},
url = {https://www.rfc-editor.org/info/rfc7464},
author = {N. Williams},
title = {{JavaScript Object Notation (JSON) Text Sequences}},
pagetotal = 8,
year = 2015,
month = feb,
} RFC text
Contents
Internet Engineering Task Force (IETF) N. Williams
Request for Comments: 7464 Cryptonector
Category: Standards Track February 2015
ISSN: 2070-1721
JavaScript Object Notation (JSON) Text Sequences
Abstract
This document describes the JavaScript Object Notation (JSON) text
sequence format and associated media type "application/json-seq". A
JSON text sequence consists of any number of JSON texts, all encoded
in UTF-8, each prefixed by an ASCII Record Separator (0x1E), and each
ending with an ASCII Line Feed character (0x0A).
Status of This Memo
This is an Internet Standards Track document.
This document is a product of the Internet Engineering Task Force
(IETF). It represents the consensus of the IETF community. It has
received public review and has been approved for publication by the
Internet Engineering Steering Group (IESG). Further information on
Internet Standards is available in Section 2 of RFC 5741.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
http://www.rfc-editor.org/info/rfc7464.
Copyright Notice
Copyright (c) 2015 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction and Motivation .....................................2
1.1. Conventions Used in This Document ..........................2
2. JSON Text Sequence Format .......................................3
2.1. JSON Text Sequence Parsing .................................3
2.2. JSON Text Sequence Encoding ................................4
2.3. Incomplete/Invalid JSON Texts Need Not Be Fatal ............4
2.4. Top-Level Values: numbers, true, false, and null ...........5
3. Security Considerations .........................................6
4. IANA Considerations .............................................6
5. Normative References ............................................7
Acknowledgements ...................................................8
Author's Address ...................................................8
1. Introduction and Motivation
The JavaScript Object Notation (JSON) [RFC7159] is a very handy
serialization format. However, when serializing a large sequence of
values as an array, or a possibly indeterminate-length or never-
ending sequence of values, JSON becomes difficult to work with.
Consider a sequence of one million values, each possibly one kilobyte
when encoded -- roughly one gigabyte. It is often desirable to
process such a dataset in an incremental manner without having to
first read all of it before beginning to produce results.
Traditionally, the way to do this with JSON is to use a "streaming"
parser, but these are not widely available, widely used, or easy to
use.
This document describes the concept and format of "JSON text
sequences", which are specifically not JSON texts themselves but are
composed of (possible) JSON texts. JSON text sequences can be parsed
(and produced) incrementally without having to have a streaming
parser (nor streaming encoder).
1.1. Conventions Used in This Document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in
[RFC2119].
2. JSON Text Sequence Format
Two different sets of ABNF rules are provided for the definition of
JSON text sequences: one for parsers and one for encoders. Having
two different sets of rules permits recovery by parsers from
sequences where some of the elements are truncated for whatever
reason. The syntax for parsers is specified in terms of octet
strings that are then interpreted as JSON texts, if possible. The
syntax for encoders, on the other hand, assumes that sequence
elements are not truncated.
JSON text sequences MUST use UTF-8 encoding; other encodings of JSON
(i.e., UTF-16 and UTF-32) MUST NOT be used.
2.1. JSON Text Sequence Parsing
The ABNF [RFC5234] for the JSON text sequence parser is as given in
Figure 1.
input-JSON-sequence = *(1*RS possible-JSON)
RS = %x1E; "record separator" (RS), see RFC 20
; Also known as: Unicode Character INFORMATION SEPARATOR
; TWO (U+001E)
possible-JSON = 1*(not-RS); attempt to parse as UTF-8-encoded
; JSON text (see RFC 7159)
not-RS = %x00-1d / %x1f-ff; any octets other than RS
Figure 1: JSON Text Sequence ABNF
In prose: a series of octet strings, each containing any octet other
than a record separator (RS) (0x1E) [RFC20]. All octet strings are
preceded by an RS byte. Each octet string in the sequence is to be
parsed as a JSON text in the UTF-8 encoding [RFC3629].
If parsing of such an octet string as a UTF-8-encoded JSON text
fails, the parser SHOULD nonetheless continue parsing the remainder
of the sequence. The parser can report such failures to
applications, which might then choose to terminate parsing of a
sequence. Multiple consecutive RS octets do not denote empty
sequence elements between them and can be ignored.
This document does not define a mechanism for reliably identifying
text sequence by position (for example, when sending individual
elements of an array as unique text sequences). For applications
where truncation is a possibility, this means that intended sequence
elements can be truncated and can even be missing entirely;
therefore, a reference to an nth element would be unreliable.
There is no end of sequence indicator.
2.2. JSON Text Sequence Encoding
The ABNF for the JSON text sequence encoder is given in Figure 2.
JSON-sequence = *(RS JSON-text LF)
RS = %x1E; see RFC 20
; Also known as: Unicode Character INFORMATION SEPARATOR
; TWO (U+001E)
LF = %x0A; "line feed" (LF), see RFC 20
JSON-text = <given by RFC 7159, using UTF-8 encoding>
Figure 2: JSON Text Sequence ABNF
In prose: any number of JSON texts, each encoded in UTF-8 [RFC3629],
each preceded by one ASCII RS character, and each followed by a line
feed (LF). Since RS is an ASCII control character, it may only
appear in JSON strings in escaped form (see [RFC7159]), and since RS
may not appear in JSON texts in any other form, RS unambiguously
delimits the start of any element in the sequence. RS is sufficient
to unambiguously delimit all top-level JSON value types other than
numbers. Following each JSON text in the sequence with an LF allows
detection of truncated JSON texts consisting of a number at the top-
level; see Section 2.4.
JSON text sequence encoders are expected to ensure that the sequence
elements are properly formed. When the JSON text sequence encoder
does the JSON text encoding, the sequence elements will naturally be
properly formed. When the JSON text sequence encoder accepts
already-encoded JSON texts, the JSON text sequence encoder ought to
parse them before adding them to a sequence.
Note that on some systems it"s possible to input RS by typing
"ctrl-^"; on some system or applications, the correct sequence may be
"ctrl-v ctrl-^". This is helpful when constructing a sequence
manually with a text editor.
2.3. Incomplete/Invalid JSON Texts Need Not Be Fatal
Per Section 2.1, JSON text sequence parsers should not abort when an
octet string contains a malformed JSON text. Instead, the JSON text
sequence parser should skip to the next RS. Such a situation may
arise in contexts where, for example, data that is appended to log
files to log files is truncated by the filesystem (e.g., due to a
crash or administrative process termination).
Incremental JSON text parsers may be used, though of course failure
to parse a given text may result after first producing some
incremental parse results.
Sequence parsers should have an option to warn about truncated JSON
texts.
2.4. Top-Level Values: numbers, true, false, and null
While objects, arrays, and strings are self-delimited in JSON texts,
numbers and the values 'true', 'false', and 'null' are not. Only
whitespace can delimit the latter four kinds of values.
JSON text sequences use 0x0A as a "canary" octet to detect
truncation.
Parsers MUST check that any JSON texts that are a top-level number,
or that might be 'true', 'false', or 'null', include JSON whitespace
(at least one byte matching the "ws" ABNF rule from [RFC7159]) after
that value; otherwise, the JSON-text may have been truncated. Note
that the LF following each JSON text matches the "ws" ABNF rule.
Parsers MUST drop JSON-text sequence elements consisting of non-self-
delimited top-level values that may have been truncated (that are not
delimited by whitespace). Parsers can report such texts as warnings
(including, optionally, the parsed text and/or the original octet
string).
For example, '<RS>123<RS>' might have been intended to carry the top-
level number 1234, but it got truncated. Similarly, '<RS>true<RS>'
might have been intended to carry the invalid text 'trueish'.
'<RS>truefalse<RS>' is not two top-level values, 'true', and 'false';
it is simply not a valid JSON text.
Implementations may produce a value when parsing '<RS>"foo"<RS>'
because their JSON text parser might be able to consume bytes
incrementally; since the JSON text in this case is a self-delimiting
top-level value, the parser can produce the result without consuming
an additional byte. Such implementations ought to skip to the next
RS byte, possibly reporting any intervening non-whitespace bytes.
Detection of truncation of non-self-delimited sequence elements
(numbers, true, false, and null) is only possible when the sequence
encoder produces or receives complete JSON texts. Implementations
where the sequence encoder is not also in charge of encoding the
individual JSON texts should ensure that those JSON texts are
complete.
3. Security Considerations
All the security considerations of JSON [RFC7159] apply. This format
provides no cryptographic integrity protection of any kind.
As usual, parsers must operate on input that is assumed to be
untrusted. This means that parsers must fail gracefully in the face
of malicious inputs.
Note that incremental JSON text parsers can produce partial results
and later indicate failure to parse the remainder of a text. A
sequence parser that uses an incremental JSON text parser might treat
a sequence like '<RS>"foo"<LF>456<LF><RS>' as a sequence of one
element ("foo"), while a sequence parser that uses a non-incremental
JSON text parser might treat the same sequence as being empty. This
effect, and texts that fail to parse and are ignored, can be used to
smuggle data past sequence parsers that don't warn about JSON text
failures.
Repeated parsing and re-encoding of a JSON text sequence can result
in the addition (or stripping) of trailing LF bytes from (to)
individual sequence element JSON texts. This can break signature
validation. JSON has no canonical form for JSON texts, therefore
neither does the JSON text sequence format.
4. IANA Considerations
The MIME media type for JSON text sequences is application/json-seq.
Type name: application
Subtype name: json-seq
Required parameters: N/A
Optional parameters: N/A
Encoding considerations: binary
Security considerations: See RFC 7464, Section 3.
Interoperability considerations: Described herein.
Published specification: RFC 7464.
Applications that use this media type:
<https://stedolan.github.io/jq>
<https://github.com/mapbox/cligj>
<https://github.com/hildjj/json-text-sequence>
Fragment identifier considerations: N/A
Additional information:
o Deprecated alias names for this type: N/A
o Magic number(s): N/A
o File extension(s): N/A
o Macintosh file type code(s): N/A
Person & email address to contact for further information:
json@ietf.org
Intended usage: COMMON
Author: Nicolas Williams (nico@cryptonector.com)
Change controller: IETF
5. Normative References
[RFC20] Cerf, V., "ASCII format for network interchange", STD 80,
RFC 20, October 1969,
<http://www.rfc-editor.org/info/rfc20>.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997,
<http://www.rfc-editor.org/info/rfc2119>.
[RFC3629] Yergeau, F., "UTF-8, a transformation format of ISO
10646", STD 63, RFC 3629, November 2003,
<http://www.rfc-editor.org/info/rfc3629>.
[RFC5234] Crocker, D., Ed. and P. Overell, "Augmented BNF for Syntax
Specifications: ABNF", STD 68, RFC 5234, January 2008,
<http://www.rfc-editor.org/info/rfc5234>.
[RFC7159] Bray, T., Ed., "The JavaScript Object Notation (JSON) Data
Interchange Format", RFC 7159, March 2014,
<http://www.rfc-editor.org/info/rfc7159>.
Acknowledgements
Phillip Hallam-Baker proposed the use of JSON text sequences for
logfiles and pointed out the need for resynchronization. Stephen
Dolan created <https://github.com/stedolan/jq>, which uses something
like JSON text sequences (with LF as the separator between texts on
output, and requiring only such whitespace as needed to disambiguate
on input). Carsten Bormann suggested the use of ASCII RS, and Joe
Hildebrand suggested the use of LF in addition to RS for
disambiguating top-level number values. Paul Hoffman shepherded the
document. Many others contributed reviews and comments on the JSON
Working Group mailing list.
Author's Address
Nicolas Williams
Cryptonector, LLC
EMail: nico@cryptonector.com
AI ELI5
On this page
§1Why JSON Needs Sequences
RFC 7159 defines JSON, a handy serialization format. But when you need to send a large series of values as one array, or a series of indeterminate length or one that never ends, JSON gets hard to work with. The RFC gives an example of one million values of roughly one kilobyte each, around one gigabyte in total, which you would like to process without reading it all first.
The usual answer is a streaming parser, but the RFC notes these are not widely available, widely used or easy to use. A JSON text sequence is the alternative. It is specifically not a JSON text itself. It is a series of pieces that are each possibly a JSON text, and it can be parsed and produced incrementally without a streaming parser or a streaming encoder.
Reading a giant list all at once is like trying to swallow a whole pizza. This format cuts the pizza into slices so you can eat one slice at a time.
§2.2How to Write a Sequence
An encoder writes any number of JSON texts. Each one is encoded in UTF-8, preceded by one ASCII Record Separator (RS, byte 0x1E) and followed by a Line Feed (LF, byte 0x0A). The ABNF is JSON-sequence = *(RS JSON-text LF). Other encodings of JSON, UTF-16 and UTF-32, MUST NOT be used.
The separator works because RS is an ASCII control character. Inside a JSON string it can only appear in escaped form, and it cannot appear in a JSON text in any other form, so RS unambiguously marks the start of every element. RS alone is enough to delimit every top-level JSON value type except numbers, and the trailing LF covers that gap (see the section on top-level values).
Encoders are expected to make sure each element is well formed. If the encoder is handed already-encoded JSON texts, it ought to parse them before adding them to the sequence. To build a sequence by hand in a text editor, some systems let you type RS as ctrl-^, and others need ctrl-v ctrl-^.
Each message gets a special stamp in front and a line break at the end. The stamp is a character that can never show up inside a message, so you always know where a new message starts.
JSON-sequence = *(RS JSON-text LF)
RS = %x1E; "record separator" (RS), see RFC 20
LF = %x0A; "line feed" (LF), see RFC 20
JSON-text = <given by RFC 7159, using UTF-8 encoding>§2.1How to Read a Sequence
The parser rules are looser than the encoder rules on purpose. The parser grammar is input-JSON-sequence = *(1*RS possible-JSON): a series of octet strings, each containing any octet other than RS, and each preceded by RS. Every octet string is then interpreted as a UTF-8 JSON text, if possible.
Several RS bytes in a row do not mean empty elements between them, and can be ignored. There is no end-of-sequence indicator. The RFC also warns that it defines no way to identify an element by position, so a reference to the nth element would be unreliable where truncation is possible, because elements can be truncated or even missing entirely.
The reader looks for the stamps, then reads whatever is between two stamps as one message. Extra stamps in a row are just ignored.
§2.3Surviving Broken Elements
If an octet string fails to parse as a UTF-8 JSON text, the parser SHOULD still continue with the rest of the sequence. It may report the failure to the application, which can choose to stop. In practice the parser skips ahead to the next RS.
The RFC's example of when this happens is log data appended to a file and then cut short by the filesystem, for instance after a crash or an administrator terminating a process. Incremental JSON parsers may be used, though a failure may come only after some partial results were already produced. Sequence parsers should also have an option to warn about truncated JSON texts.
If one message on the ribbon is torn, you do not throw the whole ribbon away. You skip to the next stamp and keep going.
§2.4Numbers, true, false and null
Objects, arrays and strings mark their own ends, but numbers and the values true, false and null do not. Only whitespace can end them. That matters because a cut-off 1234 looks like a complete 123. The RFC uses the LF byte as a canary to detect this kind of truncation.
Parsers MUST check that any top-level number, or anything that might be true, false or null, is followed by at least one JSON whitespace byte. Parsers MUST drop elements of this kind that are not delimited by whitespace, and can report them as warnings. The RFC's examples: <RS>123<RS> might have been meant to carry 1234, <RS>true<RS> might have been meant to carry the invalid text trueish, and <RS>truefalse<RS> is not two values but simply not valid JSON.
This check only works when the sequence encoder produces or receives complete JSON texts. If it does not encode the individual texts itself, it should ensure they are complete.
If a message just says 12 and the rest got cut off, it might have been 1234. The line break at the end is a sign that the message really finished.
- Self-delimitedObjects, arrays and strings mark their own end.
- Not self-delimitedNumbers, true, false and null need whitespace after them.
- LF canaryThe LF after each text counts as that whitespace.
- Parser dutyParsers MUST drop such elements if no whitespace follows.
§3Security Considerations
All the security considerations of JSON apply, and the format provides no cryptographic integrity protection of any kind. Parsers must treat input as untrusted and fail gracefully on malicious input.
Two specific risks are named. First, an incremental parser may return a partial result and only later fail, so a sequence like <RS>"foo"<LF>456<LF><RS> could be seen as one element by an incremental parser and as empty by a non-incremental one. That difference, and texts that fail to parse and are ignored, can be used to smuggle data past parsers that do not warn about failures. Second, repeatedly parsing and re-encoding a sequence can add or strip trailing LF bytes, which can break signature validation. JSON has no canonical form, so neither does this format.
A sneaky sender could hide something in a smudged message that one reader skips and another reads. And this format has no lock, so anyone could change the ribbon without it showing.
§4The Media Type
The RFC registers the media type application/json-seq. It has no required or optional parameters, its encoding consideration is binary, and its intended usage is listed as COMMON. The change controller is the IETF.
The registration also lists a few applications that use the type, naming the jq tool and two other projects by their addresses.
The format gets an official name tag, application/json-seq, so programs can tell each other what kind of data is coming.