This document describes run-time configuration of libucw-based programs using config files. For compile-time configuration, see [configure].
Terminology
Configuration items of all modules are organized into sections. The sections form a tree structure with top-level sections corresponding to program modules.
Each configuration item belongs to one of the following classes:
-
single value or a fixed-length array of values
-
variable-length array of values
-
subsection with several nested attributes
-
list of nodes, each being an instance of a subsection
-
bitmap of small integers (0..31) or fixed list of strings
-
exceptions (items with irregular syntax; however, they always appear as a sequence of strings, only the semantics differ)
Both fixed- and variable-length arrays consist of items of the same type. The basic types supported by the configuration mechanism are:
-
32-bit integer
-
64-bit integer
-
floating point number
-
IP address
-
string
-
choice (one of a fixed list of strings)
Program modules can define their own special types (such as network masks or attribute names) and decide how are they parsed.
Format of configuration files
Configuration files are text files that usually set one attribute per line, though it is possible to split one assignment into multiple lines and/or assign several attributes in one line. The basic format of an assignment command is
name value1 value2 ... valueN
or
name=value1 value2 ... valueN
The end of line means also end of a command unless it is preceded by a backslash. On the other hand, a semicolon terminates the command and another command can start after the semicolon. A hash starts a comment that lasts until the end of the line. A value can be enclosed in apostrophes or quotation marks and then it can contain spaces and/or control characters, otherwise the first space or control character denotes the end of the value. Values enclosed in quotation marks are interpreted as C-strings. For example, the following are valid assignment commands:
Database "main db\x2b"; Directory='index/'; Weights 100 20 30 \ 40 50 80 # a comment that is ignored
Numerical values can be succeeded by a unit. The following units are supported:
d=86400 k=1000 K=1024 h=3600 m=1000000 M=1048576 %=0.01 g=1000000000 G=1073741824
Attributes of a section or a list node can be set in two ways. First, you can write the name of the section or list, open a bracket, and then set the attributes inside the section. For example,
Section1 { Attr1 value1 Attr2 value2 ListNode { #creates a list and adds its first node Attr3 value3 Attr4 value4 } ListNode { Attr3=value5; Attr4=value6 } #appends a new node; this is still the same syntax }
The second possibility is using a shorter syntax when all attributes of a section are set on one line in a fixed order. The above example could be as well written as
Section1 { Attr1 value1 Attr2 value2 ListNode value3 value4 ListNode value5 value6 }
Of course, you cannot use the latter syntax when the attributes allow variable numbers of parameters. The parser of the configuration files checks this possibility.
If you want to set a single attribute in some section, you can also refer to the attribute as Section.Attribute.
Lists support several operations besides adding a new node. You just have to write a colon immediately after the attribute name, followed by the name of the operation. The following operations are supported:
List:clear # removes all nodes List:append { attr1=value1; ... } # adds a new node at the end List:prepend { attr1=value1; ... } # adds a new node at the beginning List:remove { attr1=search1 } # find a node and delete it List:edit { attr1=search1 } { attr1=value1; ... } # find a node and edit it List:after { attr1=search1 } { ... } # insert a node after a found node List:before { attr1=search1 } { ... } # insert a node before a found node List:copy { attr1=search1 } { ... } # duplicate a node and edit the copy List:reset { attr=value1; ... } # equivalent to :clear and :append
You can specify several attributes in the search condition and the nodes are tested for equality in all these attributes. In the editing commands, you can either open a second block with overridden attributes, or specify the new values using the shorter one-line syntax.
The commands :clear, :append, and :prepend are also supported by var-length arrays. The command :clear can also be used on string values. The following operations can be used on bitmaps: :set (which is equal to :append and :prepend), :remove, :clear, and :all (set all bits).
Including other files
To include another file, use the command
Include another/file
or if the file needs not to exist
OptionalInclude another/file
(Beware that this command has to be the last one on the line.)
Command-line parameters
The default configuration file (cf_def_file possibly overriden by environment variable cf_env_file) is read before the program is started. You can use a -C option to override the name of the configuration file. If you use this parameter several times, then all those files are loaded consecutively. A parameter -S can be used to execute a configuration command directly (after loading the default or specified configuration file). Example:
bin/program -Ccf/my-config -S'module.trace=2;module.logfile:clear' ...
If the program is compiled with debugging information, then one more parameter --dumpconfig is supported. It prints all parsed configuration items and exits.
All these switches must be used before any other parameters of the program.
Preprocessing
During compilation, all configuration files are pre-processed by a simple C-like preprocessor, which supports #ifdef, #ifndef, #if, #elsif, #else and #endif directives referring to compile-time configuration variables (the ones detected by configure script, you can see list of them in obj/autoconf.h). #if and #elsif can contain any Perl expression where each CONFIG_xyz configuration variable is substituted to 0 or 1 depending on its value.
The preprocessor also substitutes @VARIABLE@ by the value of the variable, which must be defined.
Caveats
Trying to access an unknown attribute causes an error, but unrecognized top-level sections are ignored. The reason is that a common config file is used for a lot of programs which recognize only their own sections.
Names of sections, attributes and choices are case-insensitive. Units are case-sensitive.