Stem Object and Cell Creation and Configuration Design Notes

All Stem cells and objects share the same API style in their constructor
methods (commonly 'new'). All parameters are passed as key/value
attributes and processed with Stem::Class::parse_args which is driven
with a simple table of field descriptions. Each field is described by an
anonymous hash with attribute/values. Each allowed field must have a
name and it also can have several optional description attributes. Here
are the supported field description attributes:

	name		The value of this is the name of this
			field. Obviously the field name is required.

	required	This is a boolean value that says this field
			is required. If it is not set in the constructor
			call, an error string is returned.

	default		The value for this is the default for this
			field. It is used if this field is not set in
			the constructor.

	class		The value for this is a Stem class (Perl
			package) name. It means the value of this
			parameter will be parsed as the arguments to the
			constructor ('new' for now) of that class. The
			object returned is saved as the value of this
			field. If the called constructor returns an
			error string, it is returned.

	class_args	The value of this is an anonymous list
			of attribute/value pairs for this class
			field. They are passed after the caller's
			arguments and so will override any duplicate
			passed in parameters.

	callback	The value of this is a code reference which is
			called to do custom construction of this
			field. The code ref is called with the new
			object, the field name and the anonymous
			reference which has the field's passed in
			values.

	type		The value for this attribute is the type of the
			field. Currently unsupported, it is meant for
			stronger type checking of parameters such as
			booleans. This will be supported soon.

	env		The value for this attribute is a name of a Stem
			environment variable. If this name is found in
			%Stem::Vars::Env then the value in that
			hash is used as the value for this
			attribute. This overrides any value passed in
			the the constructor or a default option.
			NOTE: Stem environment variables can be set from
			many places including the command line, the
			shell environment, command messages and the terminal.

Here is a typical constructor from the Stem::Log class. It has 3 fields
of which the first is required the other two have defaults. The
beginning of the constructor sub is shown below and that same two lines
of code is used in almost every class constructor.


my $field_info = [

	{
		'name'		=>	'log',
		'required'	=>	1,
	},
	{
		'name'		=>	'level',
		'default'	=>	'info',
	},
	{
		'name'		=>	'text',
		'default'	=>	'',
	},

] ;

sub new

	my $self = Stem::Class::parse_args( $field_info, @_ ) ;
	return $self unless ref $self ;


Object Creation Error Handling

Stem cells and objects are being created all the time and in many
ways. There is a standard way Stem constructor subs return errors. If the
object construction works, it returns the object. If there is an error,
it returns the error string. The caller must test constructor
returns with ref to see if they worked. This makes it easy to pass back
error messages to higher level objects with this code shown above.

The first line parses the passed arguments (in @_) with a field
description table. The second line tests if an object was created. If it
was (ref $self is true), then the constructor continues. Otherwise, the
error string is just returned to the caller. So the exact low level
error is propagated up the call tree. This is used consistently in all
of the constructors so even if you have a class which has a field which
is a class (see the 'class' field type above), and it has a parse error,
that error will be passed all the way up to the highest level call
(typically in the Stem::Config module).

Stem Cell Configuration

Stem cells are typically created by the Stem::Conf class. The primary
source of configuration data is from a file and that is handled by the
load method. Currently config files are legal Perl and are just parsed
with string eval. (Don't yell, I know it sucks but it is simple to
bootstrap.) The next revision of this module will tighten up the
specifications of config files and create a proper parser for it. The
parser will probably use Parse::RecDescent (written by our advisor
Damian Conway who will probably write the parser for us :). The config
syntax will probably be similar to what it is now, but barewords
(actually any token without whitespace) will be allowed anywhere. Only
value strings with white space will need to be quoted. Config keywords will
always be barewords. Fat comma will be supported and [] will demark
lists of anything. There won't be any hashes since this is really just a
mass of initializations and a list is fine to initialize a hash.

A Stem cell configuration is comprised of a list of attribute/value
pairs. You can also have a list of cell configurations in one file,
but each configuration is handled independently. Each configuration
entry has only a few options in the usual format of key/value pairs. The
first is 'class', which is required and it names the Stem class which
will be configured. The next one is 'name' and it is optional but almost
always used. It is the name that this cell will be registered as and
that is the address that will be used to send messages to this cell. The
last major option is 'args' and its value is a list of attribute/value
pairs used to initialize the cell.  Which set of configuration options
is what controls how a cell is created and/or registered.

The 'class' option is first processed and if it is not loaded, Stem will
load it for you. This can be done remotely which allows for a servlet
style of behavior, i.e. a request can come in and be a configuration or
converted to one and the required Stem class will be loaded and a cell
created. That cell can then be passed a message and respond to it. All
of that can occur at runtime on demand.

If the 'args' option is set, then a constructor of the class is called
and it is passed the attribute/value pairs from the list value of
'args'. The constructor method is defaulted to 'new' but that can be
overridden with the 'method' option. The constructor processes its
arguments (most likely using the Stem::Class::parse_args routine
described above) and has 3 possible return values. If it returns undef,
nothing more is done by the Stem::Conf module for this configuration. If
a string is returned, that is assumed to be an error message and it is
either printed or returned to the originator of this configuration. Any
other configurations in this file (or passed in remote configuration)
are skipped. If the retun value is a ref, then it is assumed
to be an object and it can be registered with the address in the 'name'
option.

If the 'name' option is set, that will be used to register the cell or
class itself. In most of the configuration cases, an object is created
by the class constructor with the 'args' option and it is then
registered as a cell with that name for its address. If no 'args'
option is set, then the class itself is registered under the 'name'
value and it is a class level cell. There can only be one class level
cell for any class although it could be registered under multiple names
(aliases). In addition, the value of the 'name' option is passed along
with the 'args' values to the constructor as the attribute 'reg_name'.

Here are some example classes which are configured in several of those
ways:

The Stem::Hub class needs to be initialized with a registration name
but has no other attributes. So its configuration has a 'name' and an
'args' option whose value is an empty list (that forces the constructor
to be called).

	[
		class	=>	'Stem::Hub',
		name	=>	'server',
		args	=>	[],
	]

The Stem::TTY::Msg class configuration doesn't use the 'name' option and
it used an empty 'args' value. So its constructor is called and it
returns its single class level object, and that is registered under its
class name.

	[
		class	=>	'Stem::TTY::Msg',
		args	=>	[],
	]


The Stem::Demo::Cmd class is a class level cell that just has a 'name'
option in its configuration and that is used to register the class
itself.

	[
		class	=>	'Stem::Demo::Cmd',
		name	=>	'cmd',
	]


The Stem::Sock::Msg is configured in the most common way, with 'name' and
'args' options and values for the arguments.

	[
		class	=>	'Stem::Sock::Msg',
		name	=>	'C',
		args	=>	[
			port	=> 6668,
			server	=> 1,
			data_msg	=> [
				to_cell		=> 'sw',
				to_target	=> 'c',
			],
		],
	]


Normally a single configuration file is loaded by the run_stem program
at start up time.  The Stem::Conf module also supports loading a
configuration file via a command message or another configuration (which
is similar to include files or Perl modules). A configuration which
loads a configuration file can evaluate in the current Hub or send it to
any Hub in the Stem system. This allows for centralized management of
Stem configurations. As an aid to this, the stem_msg program can be used
to send a 'load' command message to a Hub.