Although .ksy files are treated as YAML files, and YAML syntax allows quite a few representations of the same content, it is recommended to maintain a certain style in .ksy files to aid collaboration. This document serves as official .ksy style guide. In particular, we strive to make sure that all formats in our formats repository are using this style.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119.

This document is work in progress, not all sections are complete yet.

1. General formatting

MUST use:

  • Single implicit YAML document per file (i.e. no --- header).

  • No %YAML x.y version directives, no %TAG directives.

  • 2-space indent.

  • UTF-8 encoding throughout the file.

  • LF (AKA "UNIX") line endings.

  • Trailing newline character in a .ksy file.

  • Block YAML style in most general cases, unless explicitly specified/allowed otherwise.

Formatting of sequences MUST include an indent, i.e.:

Good
imports:
  - foo
  - bar
Bad
imports:
- foo
- bar

Formatting of maps-inside-sequences MUST have - delimiter and first map element on the same first line, i.e.:

Good
- id: foo
  type: u1
- id: bar
  type: u2
  enum: baz
Bad
-
  id: foo
  type: u1
-
  id: bar
  type: u2
  enum: baz

All identifiers, docstrings, comments and generally all human-readable text SHOULD be kept in English, unless there’s a very good reason not to do so.

2. Order of sections in a type spec

Use the following order of sections:

  1. meta, if present, MUST go first

  2. doc

  3. doc-ref

  4. seq

  5. instances, types, enums — use one’s best judgement to order these 3 to maximize readability

3. Meta section (meta)

Meta section is mostly used to specify additional ("meta") information for a particular file format, so 99% of time it is only present for top-level type. In some cases, meta section might be specified in intermediate types as well, for example, to switch default endianness or encoding. Thus, style recommendations for "top-level" meta sections and "intermediate" meta sections are different, as they serve slightly different purposes.

For top-level meta section, use the following order of keys:

  1. id — main identifier, MUST match the name of the .ksy file

  2. Human-readable meta-information (in order of importance, at least one MUST be present):

    1. title

    2. application — SHOULD name a particular application, if there’s any; if there are too many to list (for example, network packet formats or executables are used virtually everywhere), then one SHOULD omit this field.

    3. file-extension — if there’s only one extension, MUST be a string; if there are several, MUST be an sequence in block form and SHOULD order extensions from most popular extension to least popular.

    4. xref — keys inside MUST be in alphabetic order

    5. tags — values inside MUST be in alphabetic order

  3. Legal information — license — MUST be a valid SPDX license expression

  4. Processing instructions:

    1. ks-version — SHOULD list lowest possible KS compiler version that is able to compile this file.

    2. imports

  5. Defaults (in alphabetic order):

    1. encoding

    2. endian

For intermediate-level meta sections, use the following order of keys:

  1. Defaults (in alphabetic order):

    1. encoding

    2. endian

Note
KS syntax allows usage of some top-level elements deep inside the hierarchy — this can be useful during development, for example, for purpose of grafting one .ksy file into another quickly. However, in production-quality .ksy files, one MUST NOT use keys like title, imports or ks-version (i.e. everything except explicitly listed in a list above) on intermediate levels.

The following keys are reserved for internal use (i.e. debugging and test running) and MUST NOT be used in general-purpose .ksy files:

  • ks-debug

  • ks-opaque-types

4. Documentation

4.1. doc

Formatting:

  • Single-line documentation strings SHOULD BE formatted using raw unquoted string literals.

  • Multi-line SHOULD BE formatted using YAML literal style scalar, i.e. using : | syntax. An example:

doc: |
  File index entry contains intricate details about file in the
  archive: there are both meta-information attributes (such as file
  names, locations, various timestamps, etc) and references to
  inodes, which can be used to find file body in the container.

  For networked locations, file index entry uses an optional
  `remote_resource` type. Proper usage sequence is:

  * check `code` to be one that requires network usage
  * determine file name using `name_networked` instance and check if
    it's really a file requested by the user
  * proceed to query information from networked resource given by
    `resource` attribute

Lines should be wrapped to be 80 columns long. If it doesn’t fit into single line after wrapping, then it’s a multi-line docstring, so use proper multi-line syntax.

There is no formal conversion of docstrings into language-specific docstrings now in KS, but generally we SHOULD keep it close to CommonMark formatting, i.e.:

  • paragraphs separated by an empty line

  • bullet lists created by an asterisk * and a space at the beginning of the line

  • use backticks ` to wrap identifiers and small pieces of code

TODO: documentation contents, what should and should no be included

4.2. doc-ref

TODO

5. Sequence attributes

When specifying an attribute, one MUST use the following order of keys:

  1. Identifier(s)

    1. id

    2. -orig-id — use to specify original ID spelling if transcribing a structure from existing software and/or official spec

  2. size

  3. size-eos

  4. type

  5. Type-related keys:

    1. enum

    2. contents

    3. pad-right

    4. terminator

    5. include

    6. consume

    7. eos-error

    8. encoding

  6. process

  7. Repetition-related keys:

    1. repeat

    2. repeat-eos, repeat-expr, repeat-until

  8. if

  9. doc

  10. doc-ref

Every key is optional. Attributes SHOULD have at least id and doc — however, see below for notes about omitting id, and doc SHOULD NOT be included if it’s trivial (i.e. if it is a copy of id, and there is really nothing more to say about that attribute).

5.1. Attribute identifiers (id)

KS enforces specific identifier style in the language - lower_underscore_case (it is needed to be able to convert to other styles of identifier spelling, like UpperCamelCase or lowerCamelCase, which some target languages use).

KS allows omitting id. One MUST omit id to mark up attributes of unknown/undetermined purpose, i.e. unfinished reverse engineering work. One MUST NOT omit id to mark up reserved/unused attributes and padding, i.e. placeholder that are known to be empty and unused.

One SHOULD use the following rules to maintain consistency across various KSY files. Doing that would maintain the "principle of least surprise" and make life easier to end-users, reducing amount of guesswork.

  • For simple non-repeated fields, use a simple singular form — e.g. width, header, transaction_id, file.

  • For an array of objects (i.e. with repeat: something), use plural form — e.g. files, transactions.

  • Don’t be overly verbose: use commonly understood abbreviations liberally, if it will improve readability — e.g. src_mac or src_mac_addr instead of source_media_access_control_address

  • For fields that are designed to be used to detect file type (AKA "magic values"), use magic name, or, if there are several of them, magic1, magic2, etc.

  • For reserved fields which are known to be unused, use reserved name (or reserved1, reserved2, etc, if there are many of them)

  • For fields that designate number / count of something (in particular, number of repetitions of some other structure), use num_ prefix and plural form — i.e. num_questions, num_blocks, num_nodes

  • For fields that designate offset to some particular data structure, use ofs_ prefix and name of that data structure (as it would appear in the file) — i.e. ofs_block, ofs_queries, ofs_path

  • For fields that designate size of some particular data structure (in bytes or some other fixed units), use len_ prefix and name of that data structure — i.e. len_block (length of a single block entry), len_blocks (total length of whole blocks array, made of block entries).

Note
See Transcribing existing specs for more info on preserving / renaming identifiers when transcribing existing spec into KSY.

5.2. Trailing padding

If you’re using a size-limited substream for a structure, one MUST NOT specify manually calculated or auto-calculated extra padding to make structure consume whole substream. Just omit it — it will save memory and CPU time on parsing.

Good
seq:
  - id: header
    size: 64
    type: block
types:
  block:
    seq:
      - id: param1
        type: u4
      - id: param2
        type: u2
Bad
seq:
  - id: header
    size: 64
    type: block
types:
  block:
    seq:
      - id: param1
        type: u4
      - id: param2
        type: u2
      - id: padding
        size-eos: true
Worst
seq:
  - id: header
    size: 64
    type: block
types:
  block:
    seq:
      - id: param1
        type: u4
      - id: param2
        type: u2
      - id: padding
        size: 58
        # 64 - 4 - 2

6. Instance attributes

Instance attribute use the superset of keys which are allowed in sequence attributes (except for id), thus all ordering rules apply here as well. Keys MUST appear in this order:

  1. -orig-id (optional)

  2. io (optional)

  3. pos or value

  4. All other keys (except for id and -orig-id), in order specified in Sequence attributes

7. Transcribing existing specs

When transcribing structures already described in some other existing spec or software, note that it’s not necessary to copy existing identifiers to id keys (in verbatim or modified form) or even maintain same structures as types.

The rationale of doing so is that a lot of existing specs rely on particular standards and approaches of some target language and/or platform. Sometimes, existing specs are burdened by some legacy (i.e. they are obliged to maintain names of fields, even when its true purpose was extended since its introduction for compatibility with older software). KS, on the other hand, is cross-platform and cross-language, thus it is not necessary (and in many cases, it’s just impossible) to stick to single platform’s style. And KS-provide API is to be used by new software anyway, so you don’t usually need to be concerned with legacy compatibility.

Use -orig-id key to specify original names of fields for purposes of maintaining a reference link to parts of original spec, but, otherwise, feel free to use a more consistent and language-neutral approach in naming attributes and types.

7.1. Windows struct

For example, consider this MMCKINFO Windows structure, as specified in MSDN:

typedef struct {
  FOURCC ckid;
  DWORD  cksize;
  FOURCC fccType;
  DWORD  dwDataOffset;
  DWORD  dwFlags;
} MMCKINFO;

It is pretty inconsistent:

  • Some fields use abbreviated lower case (ckid — as it stands for "chunk ID"), others use upper camel case (dwDataOffset).

  • Some fields use so-called "Hungarian notation", i.e. prepending type information before identifier (i.e. fccType = "type, four-CC", dwDataOffset = "data offset, double word"), some don’t (ckid, cksize).

  • Some abbreviations are very brief (ckid), some are pretty verbose (dwDataOffset).

  • Actually, dwDataOffset and cksize specify offset and size of the same data structure (called "chunk’s data member" in human-readable annotation).

Also, this definition does not specify flag values (i.e. in a C struct union syntax), but instead relies of flag constant definitions elsewhere, which is also pretty inconvenient.

Recommended way to lay out that structure in KS would be something like that:

seq:
  - id: chunk_id
    -orig-id: ckid
    type: u4
    enum: four_cc
  - id: len_data
    -orig-id: cksize
    type: u4
  - id: type
    -orig-id: fccType
    type: u4
    enum: four_cc
  - id: ofs_data
    -orig-id: dwDataOffset
    type: u4
  - id: flags
    -orig-id: dwFlags
    type: flags
    size: 4
instances:
  data:
    pos: ofs_data
    size: len_data
types:
  flags:
    # add a comprehensive type that describes flags here

Note that we’ve clearly separated names and types here, used standard ofs_ and len_ prefixes for referencing offset and length of a particular structure (named "data", short for "chunk’s data member", in this case). Also, we’ve added data instance to access that structure directly.

7.2. Linux struct

Another example is ELF executable header, as specified in elf.h in Linux:

typedef struct
{
  unsigned char e_ident[EI_NIDENT];     /* Magic number and other info */
  Elf32_Half    e_type;                 /* Object file type */
  Elf32_Half    e_machine;              /* Architecture */
  Elf32_Word    e_version;              /* Object file version */
  Elf32_Addr    e_entry;                /* Entry point virtual address */
  Elf32_Off     e_phoff;                /* Program header table file offset */
  Elf32_Off     e_shoff;                /* Section header table file offset */
  Elf32_Word    e_flags;                /* Processor-specific flags */
  Elf32_Half    e_ehsize;               /* ELF header size in bytes */
  Elf32_Half    e_phentsize;            /* Program header table entry size */
  Elf32_Half    e_phnum;                /* Program header table entry count */
  Elf32_Half    e_shentsize;            /* Section header table entry size */
  Elf32_Half    e_shnum;                /* Section header table entry count */
  Elf32_Half    e_shstrndx;             /* Section header string table index */
} Elf32_Ehdr;

This one is less inconsistent, but still could be improved:

  • It uses its own convention for specifying "offset", "size" and "count" attributes.

  • It prepends e_ prefix to every element, which would serve absolutely no purpose in KS

  • It uses its own non-standard system of types (Elf32_Half, Elf32_Word, Elf32_Addr, etc)

  • e_ident actually is a complex 16-byte multi-member structure, which includes 4 bytes of magic number to identify a file format and 12 bytes worth of extra fields

  • Abbreviations are way too short (i.e. ph for "program header", sh for "section header", eh for "ELF header") for a casual user to understand its meaning without a documentation lookup. This can be easily remedied by using slightly more verbose names.

Thus, the recommended way to represent it would be:

seq:
  - id: magic
    -orig-id: e_ident
    contents:
      - 0x7f
      - "ELF"
    doc: Magic number
  # add extra members for these 12 bytes here
  - id: file_type
    -orig-id: e_type
    type: u2
    doc: Object file type
  - id: machine
    -orig-id: e_machine
    type: u2
    doc: Architecture
  # ...
  - id: ofs_program_headers
    -orig-id: e_phoff
    type: u4
    doc: Program header table file offset
  # ...
instances:
  program_headers:
    pos: ofs_program_headers
    repeat: expr
    repeat-expr: num_program_headers
    size: len_program_header
    type: program_header

7.3. C struct as "header"

Sometimes existing implementations use structures where they are actually not necessary. This is, again, frequently done to satisfy constraints of particular implementation, like C struct being of a fixed size. KS does not have these restrictions, so in some cases one can embed C struct "headers" right into the type and it would be totally ok.

For example, consider the following description:

Image file starts with a header, which consists of:

  • 4 bytes - magic number, must be 0x11335577

  • 4 bytes integer - width of image in pixels

  • 4 bytes integer - height of image in pixels

Then raw image data follows, width * height bytes.

Naive C implementation of this format would likely split this format into a "header" structure and a "body", header being declared as:

typedef struct {
    uint32_t magic;
    uint32_t width;
    uint32_t height;
} image_header_t;

Straightforward conversion of that structures would result in:

seq:
  - id: header
    type: image_header
  - id: image_data
    size: header.width * header.height
types:
  image_header:
    seq:
      - id: magic
        contents: [0x11, 0x33, 0x55, 0x77]
      - id: width
        type: u4
      - id: height
        type: u4

However, in KS, this is very redundant and unnecessarily complex. One can just put everything in one simple type — this is easier to read, understand and use (and it does not make up artificial "header" entity where we can avoid it):

seq:
  - id: magic
    contents: [0x11, 0x33, 0x55, 0x77]
  - id: width
    type: u4
  - id: height
    type: u4
  - id: image_data
    size: width * height