Getting started with Aglyph

Release:3.0.0.post1

During this brief tutorial, you will download and install Aglyph, build a simple Python application based on the MovieLister component discussed in Inversion of Control Containers and the Dependency Injection pattern, then modify the application to take advantage of Aglyph dependency injection. This process will allow you understand the Dependency Injection pattern in general, and the Aglyph approach to Dependency Injection in particular.

This tutorial is a “whirlwind tour” of Aglyph that covers only the basics. Once you have completed the steps, read the Aglyph cookbook for additional guidelines and examples. Also review the Aglyph API reference, The Aglyph Context fluent API and the Aglyph context DTD to understand the details.

The tutorial assumes that you are familiar with Python development in general, and that Python 2.7 or 3.4+ is already installed on your system.

Note

It is recommended, but not required, that you read the Inversion of Control Containers and the Dependency Injection pattern and Python Dependency Injection [PDF] articles before beginning this tutorial.

1. Download and install Aglyph

There are several options for downloading and installing Aglyph. Choose the method that best suits your needs or preferences.

Download and install a source or built distribution from SourceForge

If you use Windows, a source ZIP distribution and EXE and MSI installers are available from the Aglyph SourceForge project.

Run the EXE or MSI installer after downloading, or unpack the ZIP distribution and run the following command from within the distribution directory:

python setup.py install

Download and install a source distribution from the Python Package Index (PyPI)

The Aglyph source distribution can be downloaded from the Aglyph PyPI page.

Unpack the archive and run the following command from with the distribution directory:

python setup.py install

Clone the Aglyph repository from GitHub

To install the latest release from a clone of the Aglyph GitHub repository, execute the following commands from a shell:

git clone https://github.com/mzipay/Aglyph.git
cd Aglyph
python setup.py install

Install into a virtual environment

You can also create a Virtualenv (details not covered here) and install Aglyph into it by running the following commands from a shell (assumes the virtual environment is active):

pip install Aglyph

Regardless of installation method, verify that the installation was successful by importing the aglyph module from a Python interpreter. For example:

$ python
Python 3.5.4 (default, Oct  9 2017, 12:07:29)
[GCC 6.4.1 20170727 (Red Hat 6.4.1-1)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import aglyph
>>> aglyph.__version__
'3.0.0.post1'

2. Download, extract, and run the movielisterapp application

The sample code for this tutorial can be downloaded here (movielisterapp-basic.zip). If you don’t feel like typing everything out by hand and would prefer to just “follow along,” you can also download movielisterapp-aglyph.zip, which contains the completed tutorial source code (including the already-populated SQLite database).

Note

Both ZIP files are also available under the examples/ directory if you cloned the Aglyph GitHub repository.

Warning

Jython users will not be able to run the tutorial code because the standard Python sqlite3 module (which Jython does not support) is used by the example code.

To begin the tutorial, extract the ZIP archive to a temporary location and navigate into the application directory:

$ unzip movielisterapp-basic.zip
...
$ cd movielisterapp-basic

The movies.txt file is a simple colon-delimited text file that contains a number of title:director records, one per line:

The Colossus of Rhodes:Sergio Leone
Once Upon a Time in the West:Sergio Leone
THX 1138:George Lucas
American Graffiti:George Lucas
Once Upon a Time in America:Sergio Leone
Sixteen Candles:John Hughes
The Breakfast Club:John Hughes
Weird Science:John Hughes
Ferris Bueller's Day Off:John Hughes

This data file is read by a particular implementation of the MovieFinder class (ColonDelimitedMovieFinder), both of which can be found in the movies/finder.py module:

from movies.movie import Movie


class MovieFinder:

    def find_all(self):
        raise NotImplementedError()


class ColonDelimitedMovieFinder(MovieFinder):

    def __init__(self, filename):
        movies = []
        f = open(filename)
        for line in f:
            (title, director) = line.strip().split(':')
            movies.append(Movie(title, director))
        f.close()
        self._movies = movies

    def find_all(self):
        return self._movies

As you can see, each record is processed as a simple Movie data holder object. The movies/movie.py module holds the Movie class definition:

class Movie:

    def __init__(self, title, director):
        self.title = title
        self.director = director

Finally, we have a MovieLister class (defined in the movies/lister.py module), which uses a ColonDelimitedMovieFinder to find the movies directed by a particular director:

from movies.finder import ColonDelimitedMovieFinder


class MovieLister:

    def __init__(self):
        self._finder = ColonDelimitedMovieFinder("movies.txt")

    def movies_directed_by(self, director):
        for movie in self._finder.find_all():
            if (movie.director == director):
                yield movie

The application can be executed using the app.py script, which simply asks a MovieLister for all movies directed by Sergio Leone:

$ python app.py
The Colossus of Rhodes
Once Upon a Time in the West
Once Upon a Time in America

3. A (very) brief introduction to Dependency Injection

Examine the MovieLister class (in the movies/lister.py module) again. There are three things to note:

  1. The MovieLister class depends on a concrete implementation of MovieFinder.
  2. The ColonDelimitedMovieFinder class depends on a filename.
  3. The MovieLister is responsible for resolving both dependencies.

As a consequence of (3), neither the concrete MovieFinder implementation nor the name/location of the data file can be changed without modifying MovieLister.

In other words, it is MovieLister that controls dependency resolution. It is this aspect of control that is being inverted (“Inversion of Control”) when we talk about Dependency Injection. Rather than having MovieLister be responsible for resolving its dependencies, we instead give control to some other object (an “assembler”), which has the responsibility of injecting dependencies into MovieLister.

The dependency injection approach provides several benefits:

  • easier testing (“mock” or “stub” objects for testing are easier to manage)
  • lower general maintenance cost (e.g. the manner in which application/domain objects get initialized and connected to one another is “homogenized” in the assembler’s configuration, which makes application-wide changes easier to apply and test)
  • the separation of object configuration from object use means generally smaller and simpler application code that is focused on object behavior

Aglyph can inject dependencies using initializers – __init__ methods – or “factory” functions (type 2 “constructor” injection); or member variables, setter methods, and properties (type 3 “setter” injection).

In order to take advantage of type 2 “constructor” injection, the __init__ method or “factory” function must accept dependencies, which means we need to make some simple changes to movielisterapp

4. Make some general improvements to the movielisterapp application

As written, the basic application is somewhat change-resistant. For example, if we wish to support another implementation of MovieFinder (e.g. one that connects to a database to retrieve movie information), then we would also need to change the MovieLister implementation.

A simple solution to this problem is to change MovieLister so that it can accept a MovieFinder at initialization time:

class MovieLister:

    def __init__(self, finder):
        self._finder = finder

    def movies_directed_by(self, director):
        for movie in self._finder.find_all():
            if (movie.director == director):
                yield movie

Next, we’ll add a SQLMovieFinder class definition to the movies/finder.py module. This new implementation will use the standard Python sqlite3 module to connect to a SQLite database which stores the movies information:

import sqlite3
from movies.movie import Movie


class MovieFinder:

    def find_all(self):
        raise NotImplementedError()


class ColonDelimitedMovieFinder(MovieFinder):

    def __init__(self, filename):
        movies = []
        f = open(filename)
        for line in f:
            (title, director) = line.strip().split(':')
            movies.append(Movie(title, director))
        f.close()
        self._movies = movies

    def find_all(self):
        return self._movies


class SQLMovieFinder(MovieFinder):

    def __init__(self, dbname):
        self._db = sqlite3.connect(dbname)

    def find_all(self):
        cursor = self._db.cursor()
        movies = []
        try:
            for row in cursor.execute("select title, director from Movies"):
                (title, director) = row
                movies.append(Movie(title, director))
        finally:
            cursor.close()
        return movies

    def __del__(self):
        try:
            self._db.close()
        except:
            pass

The SQLMovieFinder expects a database name (a filename, or “:memory:” for an in-memory database). We’ll create a movies.db file so that it contains the same records as the original movies.txt file:

>>> import sqlite3
>>> conn = sqlite3.connect("movies.db")
>>> c = conn.cursor()
>>> c.execute("create table Movies (title text, director text)")
>>> for movie_fields in [("The Colossus of Rhodes", "Sergio Leone"),
...                      ("Once Upon a Time in the West", "Sergio Leone"),
...                      ("THX 1138", "George Lucas"),
...                      ("American Graffiti", "George Lucas"),
...                      ("Once Upon a Time in America", "Sergio Leone"),
...                      ("Sixteen Candles", "John Hughes"),
...                      ("The Breakfast Club", "John Hughes"),
...                      ("Weird Science", "John Hughes"),
...                      ("Ferris Bueller's Day Off", "John Hughes")]:
>>>     c.execute("insert into Movies values (?, ?)", movie_fields)
...
>>> c.close()
>>> conn.commit()
>>> conn.close()

Finally, we’ll change app.py so that the new SQLMovieFinder is used to initialize a MovieLister:

import sys

from movies.finder import SQLMovieFinder
from movies.lister import MovieLister

lister = MovieLister(SQLMovieFinder("movies.db"))
for movie in lister.movies_directed_by("Sergio Leone"):
    sys.stdout.write("%s\n" % movie.title)

Running the application again should give us the same results:

$ python app.py
The Colossus of Rhodes
Once Upon a Time in the West
Once Upon a Time in America

The basic application is now more flexible: we can change the MovieFinder implementation without having to modify the MovieLister class definition. However, we are still required to modify app.py if we decide to change the MovieFinder implementation!

Note

An important aspect of Aglyph is that it is non-intrusive, meaning that it requires only minimal changes to your existing application code in order to provide dependency injection capabilities.

Notice that the changes made in this section, while adding flexibility to the application, did not require the use of Aglyph. In fact, as we add Aglyph dependency injection support in the next two sections, no further changes to the movies/lister.py, movies/finder.py, and movies/movie.py modules need to be made.

5. Add Dependency Injection support to the movielisterapp application

Recall that Dependency Injection gives reponsibility for injecting dependencies to an an external object (called an “assembler”). In Aglyph, this “assembler” is an instance of the aglyph.assembler.Assembler class.

An aglyph.assembler.Assembler requires a “context,” which is a collection of component definitions. A component (aglyph.component.Component) is simply a description of some object, including how it is created/acquired and its dependencies. Any component can itself be a dependency of any other component(s).

In Aglyph, a context is defined by the aglyph.context.Context class. Objects of this class can be created and populated either directly or by using The Aglyph Context fluent API. A specialized subclass, aglyph.context.XMLContext, is also provided to allow a context to be defined declaratively in an XML document. Such XML documents should conform to the Aglyph context DTD.

In this section, we will create a declarative XML context and use The Aglyph Context fluent API for movielisterapp in order to demonstrate each approach.

Warning

In practice, you should choose either aglyph.context.XMLContext or aglyph.context.Context (The Aglyph Context fluent API) when configuring Aglyph for your application.

First, we’ll create the XML context document as movies-context.xml:

<?xml version="1.0" encoding="utf-8"?>
<context id="movies-context">
    <component id="delim-finder"
               dotted-name="movies.finder.ColonDelimitedMovieFinder">
        <init>
            <arg><str>movies.txt</str></arg>
        </init>
    </component>
    <component id="movies.finder.MovieFinder"
               dotted-name="movies.finder.SQLMovieFinder">
        <init>
            <arg><str>movies.db</str></arg>
        </init>
    </component>
    <component id="movies.lister.MovieLister">
        <init>
            <arg reference="movies.finder.MovieFinder" />
        </init>
    </component>
</context>

Some interesting things to note here:

  • A <context> requires an id attribute, which should uniquely identify the context.
  • A <component> requires an id attribute, and has an optional dotted-name attribute. If dotted-name is not provided, then the id attribute is assumed to be a dotted name; otherwise, the id can be a user-defined identifier and the dotted-name must be provided (this is useful when describing multiple components of the same class, for example). A dotted name is a string that represents an importable module, class, or function.
  • Initialization arguments are provided as <arg> child elements of a parent <init> element. An <arg> is a postional argument, while an <arg keyword="..."> is a keyword argument. (As in Python, the order in which positional arguments are declared is significant, while the order of keyword arguments is not.)

Note

A dotted name is a “dotted_name.NAME” or “dotted_name” string that represents a valid absolute import statement according to the following productions:

absolute_import_stmt ::=  "from" dotted_name "import" NAME
                          | "import" dotted_name
dotted_name          ::=  NAME ('.' NAME)*

Notice that the movies.lister.MovieLister component is being injected with a reference to the movies.finder.MovieFinder component, which describes an instance of movies.finder.SQLMovieFinder. We could easily change back to using movies.finder.ColonDelimitedMovieFinder by changing the reference.

Next, we’ll create an equivalent context, but this time using The Aglyph Context fluent API. In movies/__init__.py:

from movies.finder import MovieFinder, SQLMovieFinder
from movies.lister import MovieLister

from aglyph.component import Reference as ref
from aglyph.context import Context


context = Context("movies-context")

(context.component("delim-finder").
    create("movies.finder.ColonDelimitedMovieFinder").
    init("movies.txt").
    register())

# makes SQLMovieFinder the default impl bound to "movies.finder.MovieFinder"
(context.component(MovieFinder).
    create(SQLMovieFinder).
    init("movies.db").
    register())

# will initialize MovieLister with an object of SQLMovieFinder
context.component(MovieLister).init(ref(MovieFinder)).register()

Compare this context carefully with the XML declarative context above; they are identical. However, there are several interesting things to note about initializing the context using the fluent API:

  • Here we simply use the component(...) method, which results in all components being of the default type (prototype). Defining components of different types (i.e. prototype, singleton, borg, weakref) is simply a matter of using the corresponding method name. We’ll use some of these in the next part of the tutorial. These methods are the “entry points” into the fluent configuration API.
  • Each component definition is terminated by a call to the register() method. This method must be the final call, as it (a) terminates the chained-call sequence and, more importantly, (b) finalizes the compoonent definition in the context. (If you get “component not found” errors when using the fluent API, the first thing to check is that you remembered to call register()!)
  • The component methods (prototype(...) / singleton(...) / borg(...) / weakref(...)) and the create(...) method can accept dotted-name strings as well as objects. If the argument is not a string, Aglyph determines its dotted-name and uses that value. So in the above context, for example, create(SQLMovieFinder) is actually equivalent to create("movies.finder.SQLMovieFinder").
  • Unlike the component and create methods, the init(...) and set(...) (not shown here) methods do not automatically convert non-string arguments to dotted names. This is so that classes and other callables may be used directly as arguments. This is why we must use init(ref(MovieFinder)) (note the use of ref(...)) when defining the MovieLister component.

Now that we have created Aglyph configurations for movielisterapp, it’s time to modify the app.py script to use dependency injection. To demonstrate the use of both types of configution, we’ll create two different versions of the application script.

Note

As noted earlier, in practice you would choose one of the configuration options and set up your application entry point appropriately.

The app_xml.py script will use the declarative XML context:

import sys
from aglyph.assembler import Assembler
from aglyph.context import XMLContext

context = XMLContext("movies-context.xml")
assembler = Assembler(context)

lister = assembler.assemble("movies.lister.MovieLister")
for movie in lister.movies_directed_by("Sergio Leone"):
    sys.stdout.write("%s\n" % movie.title)

This script creates an assembler with a context that is read from the movies-context.xml XML document. Notice that we no longer need to create the SQLMovieFinder class directly; we have effectively separated the configuration of MovieLister from its use in the application.

Running the application produces the same results as usual:

$ python app_xml.py
The Colossus of Rhodes
Once Upon a Time in the West
Once Upon a Time in America

The app_fluent.py script will use the context that was created in movies/__init__.py:

import sys

from aglyph.assembler import Assembler
from movies import context

assembler = Assembler(context)

lister = assembler.assemble("movies.lister.MovieLister")
for movie in lister.movies_directed_by("Sergio Leone"):
    sys.stdout.write("%s\n" % movie.title)

Again, running the application produces the expected results:

$ python app_fluent.py
The Colossus of Rhodes
Once Upon a Time in the West
Once Upon a Time in America

6. Make changes to the movielisterapp application

Now that the application is configured to use Aglyph for dependency injection, let’s make some changes to demonstrate application maintenance under Aglyph.

Note

The key point of this final exercise is that we will be able to make “significant” changes to the application without having to modify any of the application source code. This is possible because we have separated the configuration of objects from their use; this is the goal of Depdendency Injection.

Introducing assembly strategies

In our existing configurations, all components are using Aglyph’s default assembly strategy, prototype, which means that each time a component is assembled, a new object is created, initialized, wired, and returned.

This is not always desired (or appropriate), so Aglyph also supports singleton, borg, and weakref assembly strategies.

For details of what each assembly strategy implies, please refer to aglyph.component.Strategy.

See also

The Borg design pattern
Alex Martelli’s original Borg recipe (from ActiveState Python Recipes)
Module weakref
Documentation of the weakref standard module.

Modify movielisterapp to use a singleton ColonDelimitedMovieFinder

We note that ColonDelimitedMovieFinder class parses its data file on every initialization. We don’t expect the data file to change very often, at least not during application runtime, so we’d prefer to only create an instance of ColonDelimitedMovieFinder once, regardless of how many times during the application runtime that it is requested (i.e. assembled). For the sake of demonstration, preted for a moment that movielisterapp is a useful application in which MovieFinder objects are used by more than just a MovieLister ;)

To accomplish this goal, we’ll modify our configurations so that the “delim-finder” component uses the singleton assembly strategy.

Also, we’ll change the movies.lister.MovieLister component so that it uses the original ColonDelimitedMovieFinder instead of SQLMovieFinder.

The modified XML context looks like this:

<?xml version="1.0" encoding="utf-8"?>
<context id="movies-context">
    <component id="delim-finder"
               dotted-name="movies.finder.ColonDelimitedMovieFinder"
               strategy="singleton">
        <init>
            <arg><str>movies.txt</str></arg>
        </init>
    </component>
    <component id="movies.finder.MovieFinder"
               dotted-name="movies.finder.SQLMovieFinder">
        <init>
            <arg><str>movies.db</str></arg>
        </init>
    </component>
    <component id="movies.lister.MovieLister">
        <init>
            <arg reference="delim-finder" />
        </init>
    </component>
</context>

We added strategy="singleton" to the “delim-finder” component, and changed the MovieLister argument to specify reference="delim-finder".

The modifed movies/__init__.py module looks like this:

from movies.finder import MovieFinder, SQLMovieFinder
from movies.lister import MovieLister

from aglyph.component import Reference as ref
from aglyph.context import Context


context = Context("movies-context")

(context.singleton("delim-finder").
    create("movies.finder.ColonDelimitedMovieFinder").
    init("movies.txt").
    register())

# makes SQLMovieFinder the default impl bound to "movies.finder.MovieFinder"
(context.borg(MovieFinder).
    create(SQLMovieFinder).
    init("movies.db").
    register())

# will initialize MovieLister with an object of ColonDelimitedMovieFinder
context.component(MovieLister).init(ref("delim-finder")).register()

We used the singleton(...) method to define the “delim-finder” component. Also, because the component ID “delim-finder” is not a dotted name, we need to manually specify that the MovieLister argument is an aglyph.component.Reference to “delim-finder”.

Running either version of the application still produces the expected results:

The Colossus of Rhodes
Once Upon a Time in the West
Once Upon a Time in America

Modify movielisterapp again to use a borg SQLMovieFinder

We also note that SQLMovieFinder doesn’t really need to create a new database connection every time it is assembled. We could use the singleton assembly strategy, but instead we’ll use a similar pattern called borg. Of course, we’ll also change the application to again use the SQLMovieFinder.

The final modified XML context looks like this:

<?xml version="1.0" encoding="utf-8"?>
<context id="movies-context">
    <component id="delim-finder"
               dotted-name="movies.finder.ColonDelimitedMovieFinder"
               strategy="singleton">
        <init>
            <arg><str>movies.txt</str></arg>
        </init>
    </component>
    <component id="movies.finder.MovieFinder"
               dotted-name="movies.finder.SQLMovieFinder"
               strategy="borg">
        <init>
            <arg><str>movies.db</str></arg>
        </init>
    </component>
    <component id="movies.lister.MovieLister">
        <init>
            <arg reference="movies.finder.MovieFinder" />
        </init>
    </component>
</context>

The final modifed movies/__init__.py looks like this:

from movies.finder import MovieFinder, SQLMovieFinder
from movies.lister import MovieLister

from aglyph.component import Reference as ref
from aglyph.context import Context


context = Context("movies-context")

(context.singleton("delim-finder").
    create("movies.finder.ColonDelimitedMovieFinder").
    init("movies.txt").
    register())

# makes SQLMovieFinder the default impl bound to "movies.finder.MovieFinder"
(context.borg(MovieFinder).
    create(SQLMovieFinder).
    init("movies.db").
    register())

# will initialize MovieLister with an object of SQLMovieFinder
context.prototype(MovieLister).init(ref(MovieFinder)).register()

Running either the app_xml.py or app_fluent.py version of the application with the final configuration changes still produces the expected results:

The Colossus of Rhodes
Once Upon a Time in the West
Once Upon a Time in America

Suggested next steps

There are many more context/configuration options available in Aglyph beyond those that have been presented in this tutorial, including support for type 2 “setter” injection using member variables, setter methods, and properties (which can also be combined with the type 3 “constructor” injection used in the movielisterapp sample application).

Suggested next steps:

  1. Read the Aglyph cookbook.
  2. Read the Aglyph API reference and The Aglyph Context fluent API.
  3. Read the Aglyph context DTD. The DTD is fully commented, and explains some of the finer points of using XML configuration.
  4. Examine the Aglyph test cases (part of the distribution; located in the tests/ directory).
  5. Start with either the movielisterapp-basic or movielisterapp-aglyph applications and make your own modifications to explore the features of Aglyph.