SPECIFICATION VERSION
0.9
STATUS
Specification is still rather in flux. Backwards compatibility is not
guaranteed between 0.9.x releases.
ABSTRACT
IOD (short for INI On Drugs) is a configuration file format that is
mostly backward-compatible with the popular INI format. It adds a few
extensions to make configuration more powerful but still lets the
configuration parseable by a regular INI parser (albeit the parse
result might differ, sometimes significantly so, but it can be exactly
the same for simple cases). An implementation can turn off some or all
of these extensions to make the configuration closer to a regular INI
file.
IOD is meant to be a general configuration format for applications.
NOTATION
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.
RATIONALE
A general configuration file format needs to be simple for computers as
well as users to read and write. Users mostly need to write
configuration, while computers need to read but increasingly nowadays
also write to configuration (for automation tasks). INI is chosen as
the basis format for several reasons: popularity, simplicity, and ease
of round-trip parsing. These features satisfy the aforementioned
requirements.
Popularity: INI format is popular on Windows as well as Unix. This
makes it easy for new users to get started with the configuration.
Simplicity: INI format is simple and very straightforward. It is
essentially assignments of key names and values, with sections.
Round-trip parsing: Round-trip parsing means preserving everything in
the file through the parsing process, including comments and formatting
(indentations, whitespaces). Most serialization format do not have
round-trip parsers: after a read and write process, original formatting
and comments are lost. Round-trip parsing means that if one loads an
INI file, modifies a key value, and saves it again, everything that is
not modified will still be the same (including whitespaces and
comments). If no keys are modified, the saved file will be identical
with the original.
Round-trip parsing is desirable in a configuration because oftentimes
valuable information is contained in the formatting/indentation
(grouping of keys) as well as comments (user explaining why she sets a
key to a certain value, dates, other notes). Software modifying a
configuration should not destroy all these.
INI vs ...
YAML. Although YAML looks nice and has many features, it lacks
round-trip parsers and has complex rules that can trip beginners or
non-programmers, e.g. significant indentation, significant whitespace
in some places (for example after colon in mapping), etc.
JSON. Although JSON is popular and simple, it lacks round-trip parsers
and some important features (e.g. comments). (Note: The documentation
for JSON Perl module mentions the phrase "round-trip", but it uses the
phrase to mean integrity of values, not preserving
comments/whitespaces.)
But note that IOD uses JSON in places.
Apache-webserver-style lacks round-trip parsers.
XML lacks round-trip parsers and is not convenient for humans to read
and write.
Why not plain INI?
First, INI format is ill-defined. There is no single standard, thus
various implementations behave differently and there are various
variants of the format. This specification intends to describe the IOD
format more precisely.
Second, INI lacks some features that I like/need, like: variable
substitution/expression, inclusion of other files, and merging between
sections. Most of these features make writing configuration less
repetitive.
SPECIFICATION
A configuration is a text file containing a sequence of lines. Encoding
MUST be UTF-8. Each line is either a blank line, a comment line, a key
line, a section line, or a directive line. Parsing is done line-by-line
and in a single pass.
Blank line
A blank line is a line containing zero or more whitespaces only. It is
ignored.
Comment line
A comment line begins with ; or # as it's first nonblank character
(note that some INI parsers do not allow indented comment; some only
recognizes ; as the comment starter character). The use of ; is
preferred.
<ws>* ";"|"#" <COMMENT>
Examples:
; this is a comment
# this is also a comment
Key line
This line sets key name and value:
<ws>* <NAME> <ws>* "=" <ws>* <VALUE> [<ws>* ";"|"#" <COMMENT>]
NAME can contain anything but newline or the equal sign (=). VALUE is
either: 1) zero or more non-newline characters with optional encoding
prefix, or a JSON string with double quotes, or a JSON array started
with "[", or a JSON hash (object) started with "{". Encoding prefix is
! followed by encoding name, followed by at least one space. Known
encodings are: j or json, h or hex, base64, e or expr, path, paths, and
none. See "Encoding" for more details about encoding.
Examples:
foo=bar
foo bar=baz
foo = bar baz ; whitespace around the equal sign will be removed
foo=!base64 YmFyIGJheg== ; bar baz
foo=!hex 00ff00 ; 3-byte binary data
foo="a JSON string\nwith newline" ; comment
foo=!j "a JSON string\nwith newline"
foo=["a json array", "because it's started", "with ["]
foo=!json [1,2,3]
foo={"a json hash":1, "because it's started":2, "with {":3}
foo=!json {"a":1,"b":2}
foo=~/logs ; will become e.g. /home/ujang/logs
foo=~budi/Pictures ; will become e.g. /home/budi/Pictures
foo="~/logs" ; a JSON-encoded string with value "~/logs"
foo=!none [ ; a single open brace character
Specifying several keys with the same name will create an array value.
Example:
a=1
a=2
will result in (in JSON):
{"a": [1,2]}
To specify expression, use the !e or !expr encoding. Example:
x=3
y=5
z=!e $x+$y ; 8
See "Expression" for more details on expressions.
Normally a key line should occur below section line, so that key
belongs to the section. But a key line is also allowed before section
line, in which it will belong to section called GLOBAL (configurable
via the parser's default_section attribute).
Section line
A section line introduces a section:
<ws>* "[" <ws>* <NAME> <ws>* "]" [<ws>* ";"|"#" <COMMENT>]
NAME is one or more non-newline characters that are not ].
Examples:
[foo]
[Section Name]
[foo.bar.baz] ; a comment
Non-contiguous sections are allowed, they will be assumed to be set/add
keys as if the section were written contiguously, e.g.:
[sect1]
a=1
[sect2]
a=1
[sect1]
a=2
b=3
will result in sect1 containing a as [1, 2] and b as 3. However, note:
[sect1]
a=1
;!merge sect1
[sect2]
d=4
[sect1]
b=2
[sect3]
c=3
sect2 will contain {"a":1, "d":4} since at the point of parsing sect2,
sect1 only contains {"a":1}. However, sect3 will contain {"a":1, "b":2,
"c":3} since at the point of parsing sect3, sect1 already becomes
{"a":1, "b":2}.
Directive line
A directive line is an unindented comment line using the ; comment
character. The comment starts with an exclamation mark (!), a directive
name (a word matching regular expression /\w+/, and zero or more
arguments separated by whitespaces. Since it is a common error to write
! instead of :!, the parser can have a configuration to allow ; to
become optional (though note that this will result in an invalid INI
file).
(";")? <ws>* "!" <ws>* <NAME> <ws>+ [ARGUMENTS]
An invalid directive will cause parsing to fail.
Examples of valid directives:
;!include somefile.iod
This directive is invalid because it uses # instead of ;:
#!include blah
This directive is invalid because it is indented:
;start of line
;!include foo
This directive is invalid because of invalid directive name:
;!include! somefile.iod
This directive is invalid because it is unknown:
;!foo
This directive is invalid because of unbalanced quotes:
;!include "somefile.ini
This directive is invalid because of missing required argument:
;!include
Argument. An argument is a sequence of one or more non-whitespace,
non-newline characters, or a JSON string. Examples:
word
argument2
yet-another-argument!
"argument as JSON string"
""
"line1\nline2"
Below is the list of known directives (<foo> signifies required
arguments, [foo] signifies optional arguments):
!include <PATH>
Include another file, as if the content of the included file is the
next lines in the current file. An included file might contain another
!include directive. If PATH is not absolute, it is assumed to be
relative to the current file. A circular include will cause the parser
to die with an error. Example:
File dir1/a.ini:
[sectionA.sub1]
a=1
;!include ../dir2/b.ini
;!include ../dir2/b3.ini
File dir2/b.ini:
b=2
;!include b2.ini
File dir2/b2.ini:
c = 3
;!include b3.ini
File dir2/b3.ini:
c=4
[sectionB]
c=1
When dir1/a.ini is parsed, the result will be (in JSON):
{
"sectionA": {
"sub1": {
"a": 1,
"b": 2,
"c": [3, 4]
},
},
"sectionB": {
"c": 1
},
}
!merge [SECTION] ...
Specify that from now on (meaning, from the current section, not the
next session), section will merge the specified section(s), in order.
The specified sections to be merged must be predeclared. To stop
merging, specify the directive without any argument.
Example:
[defaults]
d=4
[s1]
a=1
b=2
[s2]
!merge defaults s1
a=10
c=30
[s3]
[s4]
!merge
a=20
will result in (in pseudo-JSON):
{
"defaults": {"d":4},
"s1" : {"a":1, "b":2},
"s2" : {"a":10, "b":2, "c":30, "d":4},
"s3" : {"a":1, "b":2, "d":4},
"s4" : {"a":20}, // no more merging
}
!noop ...
Directive that does nothing and will be ignored. For testing.
Encoding
* json (j)
JSON encoding.
This encoding will automatically be assumed when the value starts
with " (double quote, signifying a JSON string), [ (opening brace
parenthesis, signifying a JSON array), { (opening curly brace
parenthesis, signifying a JSON hash (object).
* hex (h)
Hex encoding, e.g.:
key = !hex 48
will set key to a literal H (ASCII code for H is 0x48). Another
example:
key = !h 480a
will set key to "H\n" (a literal H followed by newline character).
* base64
Base64 encoding, e.g.:
foo = !base64 YmFyIGJheg==
will set foo to bar baz.
* expr (e)
Currently not specified.
* path
EXPERIMENTAL. A literal string, except that ~ or ~USER prefix will be
replaced with the home directory of current user, or specified user,
respectively. An extra / suffix is also allowed for directory and
will be stripped.
When an unknown user is specified, the parser should die.
For example:
log_dir = !path ~/logs
log_dir will be set to e.g /home/ujang/logs.
If the first character of the value is ~ (tilde), the path encoding
is implicitly assumed unless there is an explicit encoding
specification. So actually the previous example could just be written
as:
log_dir = ~/logs
To prevent the translation of ~, you can use !none or double quote
(JSON string):
log_dir = !none ~/logs
log_dir = "~/logs"
* paths
EXPERIMENTAL. A literal string, except that like in !path ~ or USER
prefix will be substituted with user home directory. In addition to
that, wildcard characters like *, ? will be interpreted. Wildcard
expansion is currently left to the implementation. The result will be
an array of strings.
When wildcard does not match any file (e.g. !paths foo* where there
are no files in the current directory that begins with foo), an empty
array should be returned. When an unknown user or unknown path is
given (e.g. !paths /var/zzz/* when there is no directory named
/var/zzz/) or there is an error during expansion (e.g. !paths
/root/a* when the current user does not have sufficient permission to
look into /root), the parser should die.
Example:
dist_dirs = !paths ~/repos/perl-*
Without the !paths encoding, dist_dirs will be set to a literal
string ~/repos/perl-* without the ~ translated and wildcard expanded.
With the !paths encoding, dist_dirs will be set to an array of
strings.
* none
This explicitly turns off encoding and is useful when you want to
disable implicit encoding detection. For example:
; error, unclosed JSON string
a = "
; ok, a single double-quote
b = !none "
; translated to current paths under user's home directory, e.g. "/home/ujang"
; and "/home/ujang/Pictures"
c = ~
d = ~/Pictures/
; literal "~" and "~/Pictures"
e = !none ~
f = !none ~/Pictures/
Expression
Currently expression is not fully defined and left to the
implementation. The goal is to have a simple syntax that allows
variable substitution using the $ syntax and some arithmetics/string
operations.
Unsupported features
Some INI implementations support other features, and listed below are
those unsupported by IOD, usually because the features are not popular
or too incompatible with regular INI syntax:
* Line continuation for multiline value
key=line 1 \
line 2\
line 3
Supported by Config::IniFiles. In IOD, use JSON string or encoding:
key="line 1 \nline 2\nline 3"
* Heredoc syntax for array
key=<<EOT
value1
value2
EOT
Supported by Config::IniFiles. In IOD, use multiple assignment or
JSON encoding:
key=value1
key=value2
or:
key=!j ["value1", "value2"]
GUIDELINES FOR IMPLEMENTATIONS
Implementation can provide options to turn off some features. In
general, to make an IOD configuration file not context-dependent, a
turned off feature should cause parsing to fail to notify users that
certain features are not available. A turned off feature should not
just silently makes parsing behave differently. For example, if file
inclusion is turned off, this line:
!include somefile.ini
should make parsing fail instead of continuing (without including the
file).
An exception is when an implementation provides explicit option to
ignore certain features. For example, an implementation might provide
an option to forbid expressions, or turn off expression parsing and
parse it as literal.
Below are guidelines on what parsing options an implementation can
provide:
* whether certain encodings are recognized
* whether expression is allowed
* whether the !include directive is allowed
* whether the !merge directive is allowed
* whether discontiguous section is allowed
FAQ
Summary of differencess between INI and IOD?
IOD has more features:
* various encoding for key values
Using the including base64 and variable substitution/expression.
* merge between sections
* include other files
Why are blank lines allowed in IOD?
They aid reading by human. This is in line with the goal of making it
easy for human to read. Note that many INI parsers also allow blank
lines.
What are the downsides of IOD format?
* Currently only has Perl parser (Config::IOD, Config::IOD::Reader)
INI parsers exist everywhere though, so some of the time you can
fallback to INI. Also, because the specification is simple, it is
also relatively easy to write implementations in other languages. The
default Perl implementations themselves are only about 400-600 lines
(including blank lines, not counting POD).
SEE ALSO
INI format specification, http://en.wikipedia.org/wiki/INI_file
TOML, a minimal configuration file format that looks like INI, but
diverges more, https://github.com/toml-lang/toml. This is similar in
goal and principles to IOD, with a few differences: more types and
literals (boolean, datetime), array and strings can span lines, section
can be nested using indentation. There are already implementations in
multiple languages, and a few of them are already round-tripping.