Config parsing (2/4): Proposed syntax

Russ Allbery rra at
Thu May 10 16:45:00 UTC 2001


This file documents the standardized syntax for INN configuration files.
This is the syntax that the parsing code in libinn will understand and the
syntax towards which all configuration files should move.

The basic structure of a configuration file is a tree of groups.  Each
group has a type and an optional tag, and may contain zero or more
parameter settings, an association of a name with a value.  All parameter
names and group types are simple case-sensitive strings composed of
printable ASCII characters and not containing whitespace or any of the
characters "\:;{}[]<>" or the double-quote.  A group may contain another
group (and in fact the top level of the file can be thought of as a
top-level group that isn't allowed to contain parameter settings).

Supported parameter values are booleans, integers, real numbers, strings,
and lists of strings.

The basic syntax looks like:

    group-type tag {
        parameter: value
        parameter: [ string string ... ]
        # ...

        group-type tag {
            # ...

Tags are strings, with the same syntax as a string value for a parameter;
they are optional and may be omitted.  A tag can be thought of as the name
of a particular group, whereas the <group-type> says what that group is
intended to specify and there may be many groups with the same type.

The second parameter example above has as its value a list.  The square
brackets are part of the syntax of the configuration file; lists are
enclosed in square brackets and the elements are space-separated.

As seen above, groups may be nested.

Multiple occurances of the same parameter in the parameter section of a
group is an error.  In practice, the second parameter will take precedent,
but an error will be reported when such a configuration file is parsed.

Parameter values inherit.  In other words, the structure:

    first {
        first-parameter: 1
        second {
            second-parameter: 1
            third { third-parameter: 1 }

        another "tag" { }

is parsed into a tree that looks like:

    +-------+   +--------+   +-------+
    | first |-+-| second |---| third |
    +-------+ | +--------+   +-------+
              | +---------+
              +-| another |

where each box is a group.  The type of the group is given in the box;
none of these groups have tags except for the only group of type
"another", which has the tag "tag".  The group of type "third" has three
parameters set, namely "third-parameter" (set in the group itself),
"second-parameter" (inherited from the group of type "second"), and
"first-parameter" (inherited from "first" by "second" and then from
"second" by "third").

The practical meaning of this is that enclosing groups can be used to set
default values for a set of subgroups.  For example, consider the
following configuration that defines three peers of a news server and
newsgroups they're allowed to send:

    peer { newsgroups: * }
    peer { newsgroups: * }
    peer { newsgroups: * }

This could instead be written as:

    group {
        newsgroups: *

        peer { }
        peer { }
        peer { }

or as:

    peer {
        newsgroups: *

        peer { }
        peer { }

and for a client program that only cares about the defined list of peers,
these three structures would be entirely equivalent; all questions about
what parameters are defined in the peer groups would have identical
answers either way this configuration was written.

Note that the second form above is preferred as a matter of style to the
third, since otherwise it's tempting to derive some significance from the
nesting structure of the peer groups.  Also note that in the second
example above, the enclosing group *must* have a type other than "peer";
to see why, consider the program that asks the configuration parser for a
list of all defined peer groups and uses the resulting list to build some
internal data structures.  If the enclosing group in the second example
above had been of type peer, there would be four peer groups instead of
three and one of them wouldn't have a tag, probably provoking an error

Boolean values may be given as yes, true, or on, or as no, false, or off.
Integers must be between -2,147,483,647 and +2,147,483,647 inclusive (the
same as a C99 signed long).  Floating point numbers must be between 0 and
1e37 in absolute magnitude (the same as a C99 double) and can safely
expect eight digits of precision.

Strings are optionally enclosed in double quotes, and must be quoted if
they contain any whitespace, double-quote, or any characters in the set
"\:;[]{}<>".  Escape sequences in strings (sequences beginning with \) are
parsed the same as they are in C.  Strings can be continued on multiple
lines by ending each line in a backslash.

Lists of strings are delimited by [] and consist of whitespace-separated
strings, which must follow the same quoting rules as all other strings.
Group tags are also strings and follow the same quoting rules.

There are two more bits of syntax.  Normally, parameters must be separated
by newlines, but for convenience it's possible to put multiple parameters
on the same line separated by semicolons:

    parameter: value; parameter: value

Finally, the body of a group may be defined in a separate file.  To do
this, rather than writing the body of the group enclosed in {}, instead
give the file name in <>:

    group tag <filename>

(The filename is also a string and may be double-quoted if necessary, but
since file names rarely contain any of the excluded characters it's rarely

Here is the (almost) complete ABNF for the configuration file syntax.  In
the following, CRLF represents the platform-specific line termination
character or characters; since INN runs on Unix, this will generally just
be LF.  The syntax is per RFC 2234.

First the basic syntax elements and possible parameter values:

    WHITE               = WSP / CR / LF

    boolean             = "yes" / "on" / "true" / "no" / "off" / "false"

    integer             = ["-"] 1*DIGIT

    real-number         = ["-"] 1*DIGIT "." 1*DIGIT [ "e" ["-"] 1*DIGIT ]

    non-special         = 0x21 / 0x23-39 / 0x3D / 0x3F-5A / 0x5E-7A
                               / 0x7C / 0x7E
                                ; All VCHAR except "\:;<>[]{}

    quoted-string       = DQUOTE 1*(VCHAR / 0x8A-FF) DQUOTE
                                ; DQUOTE within the quoted string must be
                                ; written as 0x5C.22 (\"), and backslash
                                ; sequences are interpreted as in C
                                ; strings.  Characters > 0x7F are
                                ; interpreted per RFC 2279 (UTF-8).

    string              = 1*non-special / quoted-string

    list-body           = string *( 1*WHITE string )

    list                = "[" *WHITE [ list-body ] *WHITE "]"

Now the general structure:

    parameter-name      = 1*non-special

    parameter-value     = boolean / integer / real-number / string / list

    parameter           = parameter-name ":" 1*WSP parameter-value

    parameter-list      = parameter [ *WHITE (";" / CRLF) *WHITE parameter ]

    group-list          = group *( *WHITE group )

    group-body          = parameter-list [ *WHITE CRLF *WHITE group-list ]
                        / group-list

    group-file          = string

    group-contents      = "{" *WHITE [ group-body ] *WHITE "}"
                        / "<" group-file ">"

    group-type          = 1*non-special

    group-tag           = string

    group               = group-type [ 1*WSP group-tag ] 1*WSP group-contents

    file                = *WHITE *( group *WHITE )

In addition, any line beginning with "#", optionally preceeded by
whitespace, is regarded as a comment and discarded before parsing.  The
line must begin with "#" (and optional whitespace); comments at the end of
lines aren't permitted.  Comments are removed prior to parsing and
therefore aren't represented in the above grammar.

Russ Allbery (rra at             <>

More information about the inn-workers mailing list