Skip to the content.

CPAN version Build status Coverage Status


Workflow - Simple, flexible system to implement workflows


This documentation describes version 1.57 of Workflow


use Workflow::Factory qw( FACTORY );

# Defines a workflow of type 'myworkflow'
my $workflow_conf  = 'workflow.xml';

# contents of 'workflow.xml'

    <time_zone>local</time_zone>                         <!-- optional -->
    <description>This is my workflow.</description>      <!-- optional -->
    <history_class>My::Workflow::History</history_class> <!-- optional -->

    <state name="INITIAL">
        <action name="upload file" resulting_state="uploaded" />
    <state name="uploaded" autorun="yes">
        <action name="verify file" resulting_state="verified file">
             <!-- everyone other than 'CWINTERS' must verify -->
             <condition test="$context->{user} ne 'CWINTERS'" />
        <action name="null" resulting_state="annotated">
             <condition test="$context->{user} eq 'CWINTERS'" />
    <state name="verified file">
        <action name="annotate">
            <condition name="can_annotate" />
        <action name="null">
            <condition name="!can_annotate" />
    <state name="annotated" autorun="yes" may_stop="yes">
        <action name="null" resulting_state="finished">
           <condition name="completed" />
    <state name="finished" />

# Defines actions available to the workflow
my $action_conf    = 'action.xml';

# contents of 'action.xml'

    <action name="upload file" class="MyApp::Action::Upload">
        <field name="path" label="File Path"
               description="Path to file" is_required="yes" />

    <action name="verify file" class="MyApp::Action::Verify">
        <validator name="filesize_cap">

    <action name="annotate"    class="MyApp::Action::Annotate" />

    <action name="null"        class="Workflow::Action::Null" />

# Defines conditions available to the workflow
my $condition_conf = 'condition.xml';

# contents of 'condition.xml'

    <condition name="can_annotate"
               class="MyApp::Condition::CanAnnotate" />

# Defines validators available to the actions
my $validator_conf = 'validator.xml';

# contents of 'validator.xml'

    <validator name="filesize_cap" class="MyApp::Validator::FileSizeCap">
        <param name="max_size" value="20M" />

# Stock the factory with the configurations; we can add more later if
# we want
    workflow   => $workflow_conf,
    action     => $action_conf,
    condition  => $condition_conf,
    validator  => $validator_conf

# Instantiate a new workflow...
my $workflow = $self->_factory()->create_workflow( 'myworkflow' );
print "Workflow ", $workflow->id, " ",
      "currently at state ", $workflow->state, "\n";

# Display available actions...
print "Available actions: ", $workflow->get_current_actions, "\n";

# Get the data needed for action 'upload file' (assumed to be
# available in the current state) and display the fieldname and
# description

print "Action 'upload file' requires the following fields:\n";
foreach my $field ( $workflow->get_action_fields( 'FOO' ) ) {
    print $field->name, ": ", $field->description,
          "(Required? ", $field->is_required, ")\n";

# Add data to the workflow context for the validators, conditions and
# actions to work with

my $context = $workflow->context;
$context->param( current_user => $user );
$context->param( sections => \@sections );
$context->param( path => $path_to_file );

# Execute one of them
$workflow->execute_action( 'upload file' );

print "New state: ", $workflow->state, "\n";

# Later.... fetch an existing workflow
my $id = get_workflow_id_from_user( ... );
my $workflow = $self->_factory()->fetch_workflow( 'myworkflow', $id );
print "Current state: ", $workflow->state, "\n";


The eg/ticket/ directory contains a configured workflow system. You can access the same data and logic in two ways:

To initialize:

    perl --db

To run the command-line application:


To access the database and data from CGI, add the relevant configuration for your web server and call ticket.cgi:

To start up the standalone web server:


(Barring changes to HTTP::Daemon and forking the standalone server won’t work on Win32; use CGI instead, although patches are always welcome.)

For more info, see eg/ticket/README



This is a standalone workflow system. It is designed to fit into your system rather than force your system to fit to it. You can save workflow information to a database or the filesystem (or a custom storage). The different components of a workflow system can be included separately as libraries to allow for maximum reusibility.

User Point of View

As a user you only see two components, plus a third which is really embedded into another:

Developer Point of View

The workflow system has four basic components:


Just a Bunch of States

A workflow is just a bunch of states with rules on how to move between them. These are known as transitions and are triggered by some sort of event. A state is just a description of object properties. You can describe a surprisingly large number of processes as a series of states and actions to move between them. The application shipped with this distribution uses a fairly common application to illustrate: the trouble ticket.

When you create a workflow you have one action available to you: create a new ticket (‘create issue’). The workflow has a state ‘INITIAL’ when it is first created, but this is just a bootstrapping exercise since the workflow must always be in some state.

The workflow action ‘create issue’ has a property ‘resulting_state’, which just means: if you execute me properly the workflow will be in the new state ‘CREATED’.

All this talk of ‘states’ and ‘transitions’ can be confusing, but just match them to what happens in real life – you move from one action to another and at each step ask: what happens next?

You create a trouble ticket: what happens next? Anyone can add comments to it and attach files to it while administrators can edit it and developers can start working on it. Adding comments does not really change what the ticket is, it just adds information. Attachments are the same, as is the admin editing the ticket.

But when someone starts work on the ticket, that is a different matter. When someone starts work they change the answer to: what happens next? Whenever the answer to that question changes, that means the workflow has changed state.

Discover Information from the Workflow

In addition to declaring what the resulting state will be from an action the action also has a number of ‘field’ properties that describe that data it required to properly execute it.

This is an example of discoverability. This workflow system is setup so you can ask it what you can do next as well as what is required to move on. So to use our ticket example we can do this, creating the workflow and asking it what actions we can execute right now:

my $wf = Workflow::$self->_factory()->create_workflow( 'Ticket' );
my @actions = $wf->get_current_actions;

We can also interrogate the workflow about what fields are necessary to execute a particular action:

print "To execute the action 'create issue' you must provide:\n\n";
my @fields = $wf->get_action_fields( 'create issue' );
foreach my $field ( @fields ) {
    print $field->name, " (Required? ", $field->is_required, ")\n",
          $field->description, "\n\n";

Provide Information to the Workflow

To allow the workflow to run into multiple environments we must have a common way to move data between your application, the workflow and the code that moves it from one state to another.

Whenever the Workflow::Factory creates a new workflow it associates the workflow with a Workflow::Context object. The context is what moves the data from your application to the workflow and the workflow actions.

For instance, the workflow has no idea what the ‘current user’ is. Not only is it unaware from an application standpoint but it does not presume to know where to get this information. So you need to tell it, and you do so through the context.

The fact that the workflow system proscribes very little means it can be used in lots of different applications and interfaces. If a system is too closely tied to an interface (like the web) then you have to create some potentially ugly hacks to create a more convenient avenue for input to your system (such as an e-mail approving a document).

The Workflow::Context object is extremely simple to use – you ask a workflow for its context and just get/set parameters on it:

# Get the username from the Apache object
my $username = $r->connection->user;

# ...set it in the context
$wf->context->param( user => $username );

# somewhere else you'll need the username:

$news_object->{created_by} = $wf->context->param( 'user' );

Controlling What Gets Executed

A typical process for executing an action is:

When you execute the action a number of checks occur. The action needs to ensure:

Once the action passes these checks and successfully executes we update the permanent workflow storage with the new state, as long as the application has declared it.



It’s useful to have your workflow generate events so that other parts of a system can see what’s going on and react. For instance, say you have a new user creation process. You want to email the records of all users who have a first name of ‘Sinead’ because you’re looking for your long-lost sister named ‘Sinead’. You’d create an observer class like:

package FindSinead;

sub update {
    my ( $class, $wf, $event, $new_state ) = @_;
    return unless ( $event eq 'state change' );
    return unless ( $new_state eq 'CREATED' );
    my $context = $wf->context;
    return unless ( $context->param( 'first_name' ) eq 'Sinead' );

    my $user = $context->param( 'user' );
    my $username = $user->username;
    my $email    = $user->email;
    my $mailer = get_mailer( ... );
    $mailer->send( '','Found her!',
                   "We found Sinead under '$username' at '$email' );

And then associate it with your workflow:

    <observer class="FindSinead" />

Every time you create/fetch a workflow the associated observers are attached to it.

Events Generated

You can attach listeners to workflows and catch events at a few points in the workflow lifecycle; these are the events fired:


You configure the observers directly in the ‘workflow’ configuration item. Each ‘observer’ may have either a ‘class’ or ‘sub’ entry within it that defines the observer’s location.

We load these classes at startup time. So if you specify an observer that doesn’t exist you see the error when the workflow system is initialized rather than the system tries to use the observer.

For instance, the following defines two observers:

  <description>This is...</description>

  <observer class="SomeObserver" />
  <observer sub="SomeOtherObserver::Functions::other_sub" />

In the first declaration we specify the class (‘SomeObserver’) that will catch observations using its update() method. In the second we’re naming exactly the subroutine (‘other_sub()’ in the class ‘SomeOtherObserver::Functions’) that will catch observations.

All configured observers get all events. It’s up to each observer to figure out what it wants to handle.


The following documentation is for the workflow object itself rather than the entire system.

Object Methods

execute_action( $action_name, $autorun )

Execute the action $action_name. Typically this changes the state of the workflow. If $action_name is not in the current state, fails one of the conditions on the action, or fails one of the validators on the action an exception is thrown. $autorun is used internally and is set to 1 if the action was executed using autorun.

After the action has been successfully executed and the workflow saved we issue a ‘execute’ observation with the old state, action name and an autorun flag as additional parameters. So if you wanted to write an observer you could create a method with the signature:

sub update {
    my ( $class, $workflow, $action, $old_state, $action_name, $autorun )
       = @_;
    if ( $action eq 'execute' ) { .... }

We also issue a ‘change state’ observation if the executed action resulted in a new state. See “WORKFLOWS ARE OBSERVABLE” above for how we use and register observers.

Returns: new state of workflow

get_current_actions( $group )

Returns a list of action names available from the current state for the given environment. So if you keep your context() the same if you call execute_action() with one of the action names you should not trigger any condition error since the action has already been screened for conditions. If you want to divide actions in groups (for example state change group, approval group, which have to be shown at different places on the page) add group property to your action

<action name=”terminate request” group=”state change” class=”MyApp::Action::Terminate” /> <action name=”approve request” group=”approval” class=”MyApp::Action::Approve” />

my @actions = $wf->get_current_actions(“approval”);

$group should be string that reperesents desired group name. In @actions you will get list of action names available from the current state for the given environment limited by group. $group is optional parameter.

Returns: list of strings representing available actions

get_action( $action_name )

Retrieves the action object associated with $action_name in the current workflow state. This will throw an exception if:

get_action_fields( $action_name )

Return a list of Workflow::InputField objects for the given $action_name. If $action_name not in the current state or not accessible by the environment an exception is thrown.

Returns: list of Workflow::InputField objects

add_history( @( \%params | $wf_history_object ) )

Adds any number of histories to the workflow, typically done by an action in execute_action() or one of the observers of that action. This history will not be saved until execute_action() is complete.

You can add a list of either hashrefs with history information in them or full Workflow::History objects. Trying to add anything else will result in an exception and none of the items being added.

Successfully adding the history objects results in a ‘add history’ observation being thrown. See “WORKFLOWS ARE OBSERVABLE” above for more.

Returns: nothing


Returns list of history objects for this workflow. Note that some may be unsaved if you call this during the execute_action() process.


Returns list of all unsaved history objects for this workflow.


Clears all transient history objects from the workflow object, not from the long-term storage.

set( $property, $value )

Method used to overwrite Class::Accessor so only certain callers can set properties caller has to be a Workflow namespace package.

Sets property to value or throws Workflow::Exception

Observer methods

add_observer( @observers )

Adds one or more observers to a Workflow instance. An observer is a function. See “notify_observers” for its calling convention.

This function is used internally by Workflow::Factory to implement observability as documented in the section “WORKFLOWS ARE OBSERVABLE”

notify_observers( @arguments )

Calls all observer functions registered through add_observer with the workflow as the first argument and @arguments as the remaining arguments:

$observer->( $wf, @arguments );

Used by various parts of the library to notify observers of workflow instance related events.


Unless otherwise noted, properties are read-only.

Configuration Properties

Some properties are set in the configuration file for each workflow. These remain static once the workflow is instantiated.


Type of workflow this is. You may have many individual workflows associated with a type or you may have many different types running in a single workflow engine.


Description (usually brief, hopefully with a URL…) of this workflow.


Workflow uses the DateTime module to create all date objects. The time_zone parameter allows you to pass a time zone value directly to the DateTime new method for all cases where Workflow needs to create a date object. See the DateTime module for acceptable values.

Dynamic Properties

You can get the following properties from any workflow object.


ID of this workflow. This will always be defined, since when the Workflow::Factory creates a new workflow it first saves it to long-term storage.


The current state of the workflow.

last_update (read-write)

Date of the workflow’s last update.

context (read-write, see below)

A Workflow::Context object associated with this workflow. This should never be undefined as the Workflow::Factory sets an empty context into the workflow when it is instantiated.

If you add a context to a workflow and one already exists, the values from the new workflow will overwrite values in the existing workflow. This is a shallow merge, so with the following:

$wf->context->param( drinks => [ 'coke', 'pepsi' ] );
my $context = Workflow::Context->new();
$context->param( drinks => [ 'beer', 'wine' ] );
$wf->context( $context );
print 'Current drinks: ', join( ', ', @{ $wf->context->param( 'drinks' ) } );

You will see:

Current drinks: beer, wine

Internal Methods

init( $id, $current_state, \%workflow_config, \@wf_states )

THIS SHOULD ONLY BE CALLED BY THE Workflow::Factory. Do not call this or the new() method yourself – you will only get an exception. Your only interface for creating and fetching workflows is through the factory.

This is called by the inherited constructor and sets the $current_state value to the property state and uses the other non-state values from \%config to set parameters via the inherited param().

_get_workflow_state( [ $state ] )

Return the Workflow::State object corresponding to $state, which defaults to the current state.

_set_workflow_state( $wf_state )

Assign the Workflow::State object $wf_state to the workflow.

_get_next_state( $action_name )

Returns the name of the next state given the action $action_name. Throws an exception if $action_name not contained in the current state.

Initial workflow history

When creating an initial Workflow::History record when creating a workflow, several fields are required.


This method returns a list of key/value pairs to add in the initial history record. The following defaults are returned:

Override this method to change the values from their defaults. E.g.

sub get_initial_history_data {
   return (
        user => 1,
        description => "none",
        action => "run"


The configuration of Workflow is done using the format of your choice, currently XML and Perl are implemented, but additional formats can be added. Please refer to Workflow::Config, for implementation details.

Configuration examples

XML configuration

    <class>My::Workflow</class>                     <!-- optional -->
    <initial_state>INITIAL</initial_state>          <!-- optional -->
    <time_zone>local</time_zone>                    <!-- optional -->
    <description>This is my workflow.</description> <!-- optional -->

    <!-- List one or more states -->
    <state name="INITIAL">
        <action name="upload file" resulting_state="uploaded" />
        <action name="cancel upload" resulting_state="finished" />

    <state name="uploaded">
        <action name="verify file">
           <resulting_state return="redo"     state="INITIAL" />
           <resulting_state return="finished" state="finished"/>

    <state name="finished" />


As of version 2.0, Workflow allows application developers to select their own logging solution of preference: The library is a Log::Any log producer. See Log::Any::Adapter for examples on how to configure logging. For those wanting to keep running their Log::Log4perl configuration, please install Log::Any::Adapter::Log4perl and add one use statement and one line after the initialization of Log::Log4perl:

use Log::Log4perl;
use Log::Any::Adapter;   # Add this additional use-statement

Log::Any::Adapter->set( 'Log4perl' ); # Additional: Log::Any initialization


The full list of dependencies is specified in the cpanfile in the distribution archive. Additional dependencies are listed by feature. The following features are currently supported by this distribution:



CPAN testers reports however do demonstrate a problem with one of the dependencies of Workflow, namely XML::Simple.

The XML::Simple makes use of Lib::XML::SAX or XML::Parser, the default.

In addition XML::Parser can make use of plugin parsers and some of these might not be able to parse the XML utilized in Workflow. This problem has been observed with XML::SAX::RTF.

The following diagnostic points to the problem:

    No _parse_* routine defined on this driver (If it is a filter, remember to
    set the Parent property. If you call the parse() method, make sure to set a
    Source. You may want to call parse_uri, parse_string or parse_file instead.)

Your XML::SAX configuration is located in the file:



Known bugs and limitations can be seen in the Github issue tracker:


Bug reporting should be done either via Github issues

A list of currently known issues can be seen via the same URL.


The test suite can be run using prove

% prove --lib

Some of the tests are reserved for the developers and are only run of the environment variable TEST_AUTHOR is set to true. Requirements for these tests will only be installed through Dist::Zilla’s authordeps command:

% dzil authordeps --missing | cpanm --notest

The test to verify the (http/https) links in the POD documentation will only run when the variable POD_LINKS is set.


Currently the code is formatted using Perl::Tidy. The resource file can be downloaded from the central repository.



The Workflow project is currently hosted on GitHub


The code is kept under revision control using Git:



Copyright (c) 2003 Chris Winters and Arvato Direct; Copyright (c) 2004-2021 Chris Winters. All rights reserved.

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.


Jonas B. (jonasbn), current maintainer.

Chris Winters, original author.

The following folks have also helped out (listed here in no particular order):

Thanks for to Michiel W. Beijen for fix to badly formatted URL, included in release 1.52

Several PRs (13 to be exact) from Erik Huelsmann resulting in release 1.49. Yet another batch of PRs resulted in release 1.50

PR from Mohammad S Anwar correcting some POD errors, included in release 1.49

Bug report from Petr Pisar resulted in release 1.48

Bug report from Tina Müller (tinita) resulted in release 1.47

Bug report from Slaven Rezić resulting in maintenance release 1.45

Feature and bug fix by dtikhonov resulting in 1.40 (first pull request on Github)

Sérgio Alves, patch to timezone handling for workflow history deserialized using DBI persister resulting in 1.38

Heiko Schlittermann for context serialization patch resulting in 1.36

Scott Harding, for lazy evaluation of conditions and for nested conditions, see Changes file: 1.35

Oliver Welter, patch implementing custom workflows, see Changes file: 1.35 and patch related to this in 1.37 and factory subclassing also in 1.35. Improvements in logging for condition validation in 1.43 and 1.44 and again a patch resulting in release 1.46

Steven van der Vegt, patch for autorun in initial state and improved exception handling for validators, see Changes file: 1.34_1

Andrew O’Brien, patch implementing dynamic reloaded of flows, see Changes file: 1.33

Sergei Vyshenski, bug reports - addressed and included in 1.33, Sergei also maintains the FreeBSD port

Alejandro Imass, improvements and clarifications, see Changes file: 1.33

Danny Sadinoff, patches to give better control of initial state and history records for workflow, see Changes file: 1.33

Thomas Erskine, for patch adding new accessors and fixing several bugs see Changes file 1.33

Ivan Paponov, for patch implementing action groups, see Changes file, 1.33

Robert Stockdale, for patch implementing dynamic names for conditions, see Changes file, 1.32

Jim Brandt, for patch to Workflow::Config::XML. See Changes file, 0.27 and 0.30

Alexander Klink, for: patches resulting in 0.23, 0.24, 0.25, 0.26 and 0.27

Michael Bell, for patch resulting in 0.22

Martin Bartosch, for bug reporting and giving the solution not even using a patch (0.19 to 0.20) and a patch resulting in 0.21

Randal Schwartz, for testing 0.18 and swiftly giving feedback (0.18 to 0.19)

Chris Brown, for a patch to Workflow::Config::Perl (0.17 to 0.18)

Dietmar Hanisch - Provided most of the good ideas for the module and an excellent example of everyday use.

Tom Moertel gave me the idea for being able to attach event listeners (observers) to the process.

Michael Roberts graciously released the ‘Workflow’ namespace on CPAN; check out his Workflow toolkit at

Michael Schwern barked via RT about a dependency problem and CPAN naming issue.

Jim Smith - Contributed patches (being able to subclass Workflow::Factory) and good ideas.

Martin Winkler - Pointed out a bug and a few other items.