The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.

NAME

Config::Model::Manual::ModelCreationIntroduction - Introduction to model creation with Config::Model

VERSION

version 2.098

Introduction

This page describes how to write a simple configuration model. Creation of more complex models are described in Creating a model with advanced features.

Note that this document shows a lot of Perl data structure to highlight the content of a model. A Perl data structure is very similar to a JSON structure. The only thing you need to know are:

  • Curly braces { ... } contain a dictionary of key, value pairs (a hash in Perl land))

  • Square brackets [ ... ] contain a list of items (array or list in Perl land)

Some definitions

configuration file

Text file where configuration data are stored. This configuration file is used by an application -- the target application

configuration tree

The semantic content of the configuration file stored in a tree representation

configuration model

Structure and constraints of the configuration tree. Like a schema for the configuration tree

target application

The application that uses the configuration file. The application can be of type system (i.e. the configuration file is located in /etc), user (i.e. the configuration file is located in a user directory like ~/.config) or application (the configuration file is in or below the current directory)

end user

User of the target application

application developer

Target application developer

model developer

People developing the configuration model. Not necessarily the application developer

What is a configuration tree?

Most configuration files are actually organized mostly as a tree structure. Depending on the syntax of the file, this structure may be obvious to see (e.g. for XML, Apache) or not so obvious (Xorg syntax, INI syntax).

For some files like approx.conf or adduser.conf, this tree structure is quite flat. It looks much like a rake than a tree, but still, it's a tree.

For instance, this approx.conf:

 $pdiffs     1
 $max_wait   14
 debian     http://ftp.fr.debian.org/debian

can have this tree representation:

 root
 |--pdiff=1
 |--max_wait=14
 `--distrib(debian)=http://ftp.fr.debian.org/debian

Other configuration files like apache2.conf or xorg.conf have a structure that look more like a tree.

For instance, consider this xorg.conf snippet:

 Section "Device"
    Identifier     "Device0"
    Driver         "nvidia"
 EndSection
 
 Section "Screen"
    Identifier     "Screen0"
    Device         "Device0"
    Option         "AllowGLXWithComposite" "True"
    Option         "DynamicTwinView" "True"
    SubSection     "Display"
        Depth       24
    EndSubSection
 EndSection

Knowing that Xorg.conf can have several Device or Screen sections identified by their Identifiers, the configuration can be represented in this tree as:

 root
 |--Device(Device0)
 |  `--Driver=nvidia
 `--Screen(Screen0)
    |--Device=Device0
    |--Option
    |  |--AllowGLXWithComposite=True
    |  `--DynamicTwinView=True
    `--Display
       `--Depth=24

One may argue that some Xorg parameter refer to others (i.e.Device and Monitor value in Screen section) and so they cannot be represented as a tree. That's right, there are some more complex relations that are added to the tree structure. This will be covered in more details when dealing with complex models.

In some other case, the structure of a tree is not fixed. For instance, Device options in Xorg.conf are different depending on the value of the Device Driver. In this case, the structure of the configuration tree must be adapted (morphed) depending on a parameter value.

Just like XML data can have Schema to validate their content, the configuration tree structure needs to have its own schema to validate its content. Since the tree structure cannot be represented as a static tree without reference, XML like schema are not enough to validate configuration data.

Config::Model provides a kind of schema for configuration data that takes care of the cross references mentioned above and of the dynamic nature of the configuration tree required for Xorg (and others).

What is a model?

A configuration model defines the configuration tree structure:

  • A model defines one or more configuration class

  • At least one class is required to define the configuration tree root

  • Each class contains several elements. An element can be:

    • A leaf to represent one configuration parameter

    • A list of hash of leaves to represent several parameter

    • A node to hold a node of a configuration tree

    • A list or hash of nodes

These basic relations enable to define the main parts of a configuration tree.

If we refer to the approx.conf example mentioned above, one only class is required (let's say the Approx class). This class must contain (see approx.conf man page):

  • A boolean leaf for pdiff (1 if not specified)

  • An integer leaf for max_wait (10 seconds unless specified otherwise)

  • A hash of string leaves for distrib (no default).

A configuration model is stored this way by Config::Model:

 {
  name => 'Approx',
  element => [
      pdiffs => {
          type => 'leaf',
          value_type => 'boolean',
          upstream_default => '1'
      },
      max_wait => {
          type => 'leaf',
          value_type => 'integer',
          upstream_default => '10'
      },
      distributions'=> {
          type => 'hash',
          index_type => 'string' ,
          cargo => {
              value_type => 'uniline',
              type => 'leaf',
          },
      }
   ]
 }

The Xorg example leads to a slightly more complex model with several classes:

  • Xorg (root class)

  • Xorg::Device

  • Xorg::Screen

  • Xorg::Screen::Option for the Screen options

  • Xorg::Screen::Display for theDisplay subsection

The root class is declared this way:

 {
  name => 'Xorg',
  element => [
              Device => {
                         type => 'hash',
                         index_type => 'string'
                         cargo => {
                                    type => 'node',
                                    config_class_name => 'Xorg::Device'
                                  },
                        },
              Screen => {
                         type => 'hash',
                         index_type => 'string'
                         cargo => {
                                   type => 'node',
                                   config_class_name => 'Xorg::Screen'
                                  },
                        },
           ]
 }

TheXorg::Screen class is:

 {
  name => 'Xorg::Screen',
  element => [
               Device => {
                           type' => 'leaf',
                           value_type => 'uniline',
                         },
               Display => {
                            type => 'hash',
                            index_type => 'integer'
                            cargo => {
                                       type => 'node',
                                       config_class_name => 'Xorg::Screen::Display'
                                     },
                          }
              Option => {
                          type => 'node',
                          config_class_name => 'Xorg::Screen::Option'
                        },
              ]
  }

It's now time to detail how the elements of a class are constructed.

Model analysis

To define the required configuration classes, you should read the documentation of the target application to :

  • Find the structure of the configuration tree

  • Identify configuration parameters, their constraints and relations

Last but not least, you should also find several valid examples of your application configuration. These examples can be used as non-regression tests and to verify that the application documentation was understood.

Model declaration

Configuration class declaration

Since writing the data structure shown below is not fun (even with Perl), you are encouraged to use the model editor provided by cme using cme meta edit command (provided by Config::Model::Itself). This commands provides a GUI to create or update your model.

When saving, cme writes the data structure in the correct directory.

Configuration class declaration (the hard way)

In summary, configuration documentation is translated in a format usable by Config::Model:

  • The structure is translated into configuration classes

  • Configuration parameters are translated into elements

  • Constraints are translated into element attributes

All models files must be written in a specific directory. For instance, for model Xorg, you must create ./lib/Config/Model/models/Xorg.pl. Other classes like Xorg::Screen can be stored in their own file ./lib/Config/Model/models/Xorg/Screen.pl or included in Xorg.pl

A model file is a Perl file containing an array for hash ref. Each Hash ref contains a class declaration:

 [ { name => 'Xorg', ... } , { name => 'Xorg::Screen', ... } ] ;

A class can have the following parameters:

  • name: mandatory name of the class

  • class_description: Description of the configuration class.

  • generated_by: Mention with a descriptive string if this class was generated by a program. This parameter is currently reserved for Config::Model::Itself model editor.

  • include: Include element description from another class.

For more details, see "Configuration_Model" in Config::Model.

For instance:

 $ cat lib/Config/Model/models/Xorg.pl
 [
   {
     name => 'Xorg',
     class_description => 'Top level Xorg configuration.',
     include => [ 'Xorg::ConfigDir'],
     element => [
                 Files => {
                           type => 'node',
                           description => 'File pathnames',
                           config_class_name => 'Xorg::Files'
                          },
                 # snip
                ]
   },
   {
     name => 'Xorg::DRI',
     element => [
                 Mode => {
                          type => 'leaf',
                          value_type => 'uniline',
                          description => 'DRI mode, usually set to 0666'
                         }
                ]
   }
 ];

Common attributes for all elements

This first set of attributes helps the user by providing guidance (with level and status) and documentation (summary and description).

All elements (simple or complex) can have the following attributes:

  • description: full length description of the attribute

  • summary: one line summary of the above description

  • level: is important, normal or hidden. The level is used to set how configuration data is presented to the user in browsing mode. Important elements are shown to the user no matter what. hidden elements are explained with the warp notion.

  • status: is obsolete, deprecated or standard (default). Warnings are shown when using a deprecated element and an exception is raised when an obsolete element is used.

See "Configuration_class" in Config::Model for details.

Leaf elements

Leaf element is the most common type to represent configuration data. A leaf element represents a specific configuration parameter.

In more details, a leaf element have the following attributes (See "Value_model_declaration" in Config::Model::Value doc):

type

Set to leaf (mandatory)

value_type

Either boolean, integer, number, enum, string, uniline (i.e. a string without "\n") (mandatory)

min

Minimum value (for integer or number)

max

Maximum value (for integer or number)

choice

Possible values for an enum

mandatory

Whether the value is mandatory or not

default

Default value that must be written in the configuration file

upstream_default

Default value that is known by the target application and thus does not need to be written in the configuration file.

To know which attributes to use, you should read the documentation of the target application.

For instance, AddressFamily parameter (sshd_config(5)) is specified with: Specifies which address family should be used by sshd(8). Valid arguments are "any", "inet" (use IPv4 only), or "inet6" (use IPv6 only). The default is "any".

For Config::Model, AddressFamily is a type leaf element, value_type enum and the application falls back to any if this parameter is left blank in sshd_config file.

Thus the model of this element is :

 AddressFamily => {
   type             => 'leaf',
   value_type       => 'enum',
   upstream_default => 'any',
   description      => 'Specifies which address family should be used by sshd(8).',
   choice           => [ 'any', 'inet', 'inet6' ]
 }

Simple list or hash element

Some configuration parameters are in fact a list or a hash of parameters. For instance, approx.conf can feature a list of remote repositories:

 # remote repositories
 debian     http://ftp.fr.debian.org/debian
 multimedia http://www.debian-multimedia.org

These repositorie URLs must be stored as a hash where the key is debian or multimedia and the associated value is a URL. But this hash must have something which is not explicit in approx.conf file: a parameter name. Approx man page mentions that: The name/value pairs [not beginning with '$' are used to map distribution names to remote repositories.. So let's use distribution as a parameter name.

The example is stored this way in the configuration tree:

 root
 |--distribution(debian)=http://ftp.fr.debian.org/debian
 `--distribution(multimedia)=http://www.debian-multimedia.org

The model needs to declare that distribution is:

  • a type hash parameter

  • the hash key is a string

  • the values of the hash are of type leaf and value_type uniline

 distribution => {
                   type => 'hash',
                   index_type => 'string',
                   cargo => {
                              type => 'leaf',
                              value_type => 'uniline',
                            },
                   summary => 'remote repositories',
                   description => 'The other name/value pairs are ...',
                 }

For more details on list and hash elements, see hash or list model declaration man page.

node element

A node element is necessary if the configuration file has more than a list of variable. In this case, the tree is deeper than a rake and a node element if necessary to provide a new node within the tree.

In the Xorg example above, the options of Xorg::Screen need their own sub-branch in the tree:

 Screen(Screen0)
   `--Option
      |--AllowGLXWithComposite=True
      `--DynamicTwinView=True

For this, a new dedicated class is necessary>Xorg::Screen::Option> (see its declaration above). This new class must be tied to the Screen class with a node element.

A node element has the following parameters:

  • type (set to node)

  • the name of the configuration class name (>config_class_name>)

So the Option node element is declared with:

 Option => {
             type => 'node',
             config_class_name => 'Xorg::Screen::Option'
           },

Hash or list of nodes

Some configuration files can feature a set of rather complex configuration entities. For instance Xorg.pl can feature several Screen or Device definitions. These definitions are identified by the Identifier parameter:

 Section "Device"
   Identifier     "Device0"
   Driver         "nvidia"
   BusID          "PCI:3:0:1"
 EndSection
 
 Section "Screen"
   Identifier     "Screen0"
   Device         "Device0"
   DefaultDepth    24
 EndSection

The Xorg configuration tree features 2 elements (Screen and Device) that use the Identifier parameters as hash keys:

 root
 |--Device(Device0)
 |  |--Driver=nvidia
 |  `--BusId=PCI:3:0:1
 `--Screen(Screen0)
    |--Device=Device0
    `--DefaultDepth=24

And the Xorg model must define these 2 parameters as hash. The cargo of this hash is of type node and refers to 2 different configuration classes, one for Device (Xorg::Device) and one for Screen (Xorg::Screen):

 {
 name => 'Xorg',
 element => [
             Device => {
                        type => 'hash',
                        index_type => 'string'
                        cargo => {
                                   type => 'node',
                                   config_class_name => 'Xorg::Device'
                                 },
                       },
             Screen => {
                        type => 'hash',
                        index_type => 'string'
                        cargo => {
                                  type => 'node',
                                  config_class_name => 'Xorg::Screen'
                                 },
                       },
          ]
 }

Configuration wizard

Both Perl/Tk and Curses interfaces feature a configuration wizard generated from a configuration model.

The wizard works by exploring the configuration tree and stopping on each important element and on each error (mostly missing mandatory parameter).

When designing a model, you have to ponder for each element:

  • The importance level of the parameter (important, normal or hidden). level is used to set how configuration data is presented to the user in wizard and browsing mode. Important elements are shown in the wizard. hidden elements are explained with the warp notion in Creating a model with advanced features.

Reading configuration files

Once the model is specified, Config::Model can generate a nice user interface, but there's still no way to load or write the configuration file.

For Config::Model to read the file, the model designer must declare in the model how to read the file (the read backend).

The read method can use one or more of the following mechanisms:

  • Built-in, e.g Perl file, INI file...

  • A plugin, i.e. a Perl Config::Model::Backend::* class like Config::Model::Backend::Augeas

  • A custom class where a read call-back must be provided

For more details, see Config::Model::BackendMgr.

The name of the backend parameter must be specified in all cases.

Using built-in read mechanism

Config::Model comes with 3 read/write built in mechanisms:

perl_file

A perl data structure (like the ones produced by Data::Dumper). See Config::Model::DumpAsData for details.

ini_file

Windows INI file (note that only simple tree structure can use this backend)

cds_file

Config::Model own serialization format (a bit like YAML). See Config::Model::Dumper for details.

With the backend name, the following parameters must be defined:

config_dir

The configuration directory

file

Config file name (optional). defaults to <config_class_name>.[pl|ini|cds]

   read_config  => [ { backend    => 'cds_file' , 
                       config_dir => '/etc/cfg_dir',
                       file       => 'cfg_file.cds', # optional
                   } ],

See "Built-in_backend" in Config::Model::BackendMgr.pm for details

Note that these parameters can also be set with the graphical configuration model editor.

Using a plugin read mechanism

A plugin backend class can also be specified with:

  read_config  => [ { backend    => 'foo' , 
                      config_dir => '/etc/cfg_dir'
                  } ]

In this case, Config::Model tries to load Config::Model::Backend::Foo. (The class name is constructed with ucfirst($backend_name))

read_config can also have custom parameters that are passed verbatim to Config::Model::Backend::Foo methods:

  read_config  => [ { backend    => 'foo' , 
                      config_dir => '/etc/cfg_dir',
                      my_param   => 'my_value',
                  } ]

This Config::Model::Backend::Foo class is expected to provide the following methods:

new
read
write

Their signatures are explained in Config::Model::BackendMgr doc on plugin backends

Using a custom class

In case the plugin mechanism is not possible, a class with an arbitrary name can be specified:

    read_config  => [ { backend => 'custom' , 
                        class => 'MyRead',
                        config_dir => '/etc/foo', # optional
                        file => 'foo.conf',       # optional
                    } ]

Even the read method can have an arbitrary name by specifying a function parameters.

For more details on available parameters on custom backends, see Config::Model::BackendMgr doc on custom backends

Using several read mechanisms

Several read mechanism can be specified to enable:

  • Migration from one syntax to another

  • Usage of different libraries (e.g. Augeas or pure Perl backend)

For instance, to try Augeas and fall back on a custom class in case of problem, specify:

  read_config => [ {
                     save => 'backup',
                     file => 'sshd_config',
                     backend => 'augeas',
                     config_dir => '/etc/ssh'
                   },
                   {
                     function => 'sshd_read',
                     backend => 'custom',
                     class => 'Config::Model::OpenSsh',
                     config_dir => '/etc/ssh'
                 } ],

Both specifications are tried in order. If Augeas backend fails (e.g. Augeas is not installed), the custom backend is used.

An exception is raised if both methods fails. This behavior is correct for OpenSsh, but it can be a problem if you want to use Config::Model to create a configuration file from scratch. In this case you should also specify the auto_create parameter:

 read_config => [ { backend => 'custom' , 
                    class => 'ProcessRead' ,
                    config_dir => '/etc/foo',
                    file  => 'foo.conf',
                    auto_create => 1,
                } ],

Writing configuration files

Read and write specifications were designed to be very similar. Most of the times, the read and write specification are identical. In this case, there's no need to enter them: the data specified in the read specification is used to write the configuration file.

Here's an example:

  write_config => [ { backend => 'custom', 
                      class => 'NewFormat' 
                      function => 'my_write',
                    } 
                  ],

Several write specification can be used. They are tried in order, until the first succeeds.

For more information, see write specification doc

Syntax migration example

By combining multiple read specification with 'one' write specification, a configuration file can be migrated from old to new syntax. The following example migrates a configuration file from a custom syntax to a perl data file:

 { 
  name => 'Example',
  element => [ ... ] ,
  read_config  => [ { backend => 'perl_file', 
                      config_dir => '/etc/my_cfg/' 
                    } , 
                    { backend => 'custom', 
                      class => 'Bar' 
                    }, 
                  ],
  write_config => [ { backend => 'perl_file', 
                      config_dir => '/etc/my_cfg/' 
                    }
                  ],
 }

How does this work ? Here's the sequence:

  1. Configuration is stored in old file /etc/my_cfg/bar.conf

  2. Config::Model tries to read the config with perl_file read backend and looks for /etc/my_cfg/example.pl. This file is not found so the read fails.

  3. Config::Model tries the second backend which succeeds and load configuration data in the configuration tree

  4. Config::Model writes data back from configuration tree using write_config backend which writes /etc/my_cfg/example.pl

  5. At the next invocation, the first read backend will successfully read the perl configuration file. The old file is left alone and can be removed later by the system admin.

Thanks to this mechanism, this operation is idempotent so it can safely be scripted in package scriplets.

SEE ALSO

Feedback welcome

Feel free to send comments and suggestion about this page at

 config-model-users at lists dot sourceforge dot net.

AUTHORS

Dominique Dumont <ddumont at cpan.org>

AUTHOR

Dominique Dumont

COPYRIGHT AND LICENSE

This software is Copyright (c) 2005-2017 by Dominique Dumont.

This is free software, licensed under:

  The GNU Lesser General Public License, Version 2.1, February 1999