Writing Software Patterns
I've spent a lot of my writing energy writing patterns. From time to time I get asked questions about why I do that and what makes a good pattern. This is a brief article about how I look at patterns with my suggestions for people who are interested in writing patterns themselves.
01 August 2006
- What is a Pattern
- Why are Patterns important?
- Important Parts of Patterns
- Common Pattern Forms
- Choosing Your Pattern Form
- Common Issues
When you look at multiple software systems, you often find recognize similarities. A set of program elements work together in roughly the same way in lots of different places, even if masked by different names and incidental variations in behavior. Experienced programmers learn how to solve common problem in particular ways, and make rough copies from what they've seen before, while adapting these copies to better fit their new home.
If possible, we'd like to capture these common solutions as libraries or frameworks. But often the variation is big enough that it's difficult to represent it as a single library. Worse still, we may find we wish to copy a solution from a system written in an entirely different programming language.
Therefore: in the early 1990's, a bunch of people in the software world developed the idea of software patterns to capture these common solutions. By writing them down, in a somewhat structured format, we can better share this otherwise implicit knowledge. And, in contrast to libraries, this writing can also explain when the solution is appropriate and what signs lead to an alternative approach
What is a Pattern
A common definition of a pattern is that it is “a solution to a problem in a context”. That’s a definition that’s always struck me as being rather unhelpful.
For me a pattern is primarily a way to chunk up advice about a topic. Chunking is important because there's such a huge amount of knowledge you need to write software. As a result there needs be ways to divide knowledge up so you don't need to remember it all - what you need is to be able to get at a particular chunk of knowledge when you need it. Only then do you need details.
The solution provides a useful focus for the chunking. With some young eager programmer asking some grizzly veteran (i.e. anyone over thirty) how to deal with a particular situation and hear the veteran say "oh - you'll need an identity map there". The colleague can then look up identity map in some suitable patterns book.
So to make this chunking work each pattern should name a solution. This solution should be concrete, at least at the level of discussion we are talking about. You should be able to go away and use the pattern once you're given the reference. If you're successful the name should enter the vocabulary of the profession. It can take a while to do this, but when you say “proxy” any reasonable professional should know what you mean.
Patterns should have recurrence, which means the solution must be applicable in lots of different situations. If you are talking about something that's a one-off, then it's not worth adding the name to the profession's vocabulary.
One of the interesting things here is that a singular solution can often lead to a recurrent pattern. This usually crops up when you see two different singular solutions which look completely different on the surface, yet have a deeper similarity - what Christopher Alexander refers to as the “core of solution”.
Let me give an example for this. I was looking at one of our
early Java web projects. On this project the team wasn't allowed to
use JSPs. So they wrote a set of Java classes which walked through a
structure of domain objects, and produced the appropriate HTML for a
particular domain object. They noticed they were getting duplication
in the code for spitting out common HTML structures for fields,
tables, etc. So they pulled all of the HTML spitting code out into a
second utility class that had methods like
label). When they did this they noticed that they could make
drastic changes to the entire web application's appearance just by
altering code in the utility class.
Later on I saw a different project. They were using XSLT to turn XML into HTML pages. But they needed to support multiple organizations who wanted the same data displayed in their own format. So they split the transformation into two steps, first producing an intermediate XML with elements like field and table, with second stage actually producing the HTML. They would have a different second stage for each organization.
Although it seems obvious as I write it now, but when I first saw these two projects I sensed there was something similar in their approaches. However it took me several months to understand the key point - splitting a transformation into two steps: logical page and physical (HTML) page. This is the “core of the solution” which I wrote up as Two Step View. One of the great intellectual challenges of patterns is finding and isolating this core amongst all the surrounding stuff that's needed on real projects.
Patterns versus Recipes
A popular, and very effective, form of technical writing is the cookbook style (eg The Perl Cookbook, Rails Recipes). There is a lot of similarity between cookbooks and patterns books. Both emphasize a problem-solution style.
I see the big difference between the two in the notion of building a vocabulary. Recipes tend to be more particular, usually tied to a particular programming language and platform. Even when patterns are tied to a platform, they try to describe more general concepts.
As a consequence of this recipes have a stronger problem focus than the solution focus in patterns.
Although my writing interest is in patterns, this reflects my interest in general design principles rather than a judgment on the relative usefulness of the two styles. Both are effective for same basic reason - they chunk based on a concrete thing somebody wants to get done today. As a result I find both very effective. You can also learn great principles from them, but it's the answers to particular questions that bring you to the table.
Why are Patterns important?
One of the quotes that I find particularly appealing when I think about the need for patterns that that part of interest in patterns came from “...observations that projects fail despite the latest technology for lack of ordinary solutions” [PLoPD 1]. Patterns provide a way to organize and name those ordinary solutions to make it easier for people to use them.
Since these solutions are ordinary, it's common that experts in a field won't find anything new in a patterns book. For such people the biggest value of a patterns book is to help them to pass on the solutions to their colleagues.
Despite my liking for patterns, I don't think that patterns are the right approach for all situations. Even in my own latest patterns book, I used a mixture of patterns and narrative text. I think the patterns helped focus the narrative and provided a good way for me to separate the details of the solutions from the overview discussion of them. Patterns are a communication medium, and like any communication technique there are situations where they work well and those where they work badly. Practice and familiarity help you tell the difference.
Important Parts of Patterns
Anyone who takes a look at patterns is usually struck by the fact that most patterns are written using a regular form. Once you look at two sets of patterns you realize hardly ever do two patterns authors use the same form. The different patterns forms all have particular qualities to them, and any patterns author will tend to pick a form that works well with their innate preferences.
Despite the various forms, most patterns do have common elements. I'll talk about the different forms later on, but I think that's easier to do if I cover some general principles first.
Patterns are Solutions
Almost anything written about patterns has a definition that reads something like “a pattern is a solution to a problem”. While I don't disagree with that statement, I think it does tend to under-emphasize the point that patterns are primarily about solutions.
I think it's important to say this, because there is a certain mystique that tends to embroil patterns. To cut through that mystique, we must never forget that the whole point of writing a pattern is to describe a recurring and useful solution. Success is all about doing that in a way that others can replicate that solution when it's appropriate. Everything else is secondary - which means that however we choose to write the pattern, whatever form we take - all has to support this. Too often I've seen patterns writers, including myself, get lost in a particular format and lose sight of this simple priority. So whenever writing is getting difficult, remember it's the solution that counts.
And the problem? Well any solution is a solution to a problem. How could you have a solution without a corresponding problem? Understanding the problem (or problems, a pattern can solve more than one) is a key part of understanding the solution. Thinking about the problem helps you focus on the “core of the solution”. It also helps us avoid sliding into too far of a tools-oriented discussion. So understanding the problem is important - indeed vital. But the solution should remain the focus of the pattern.
An Evocative Name
One of the valuable features of patterns work is that it develops a vocabulary with which we can talk about how to do things. By naming recurring solutions, we can gradually build a vocabulary of software design that goes beyond the usual issues of technology that we normally wrestle with. I understand Java's listeners, and .NET's delegates better when I know they are part of ways of implementing the Observer pattern. The name “observer” gives me a hook that gives me a way into understanding new technological concepts - which often have different names in different technologies.
Choosing good names is hard, and I've found that I'm constantly fiddling with names even very close to deadlines. Since they are going to be a vocabulary, it's worth putting a lot of effort into getting good names. Just think of what that wrinkled veteran needs to say.
As a result names should be short, but of course it's hard to come with tight names. Mine tend to be two or three words - as I feel that most of the good one word ones are taken.
If I see patterns that are alternative ways of doing something, then I like to use different adjectives that modify a common noun. So Page Controller and Front Controller are two different forms of Controller. Strictly they are forms of Input Controller in P of EAA, but I only use three words if I really have to (as I felt I did with Single Table Inheritance and it's alternatives.)
Why as well as how
When we talk about solutions, it's easy to focus on the solution itself and how to apply the solution. It's harder to talk about when the solution is appropriate and what conditions are suitable for it, or not. This is why patterns writers put emphasis on the problem, because that focuses our minds on the trigger for the pattern. It's also why patterns writers talk about forces, because forces are a way of exploring the indications and contra-indications for the pattern.
Whenever I think I have a pattern, I try to think about when I would not use the pattern. This often leads me to alternative patterns, which is why my patterns often come in groups of alternatives.
I'm particularly suspicious of whole pattern languages that only describe one set of alternatives. One of the triggers for P of EAA was my annoyance with people who talked about The One Architecture to use for J2EE. Software systems, even within a particular area such as enterprise applications, live in a diverse world. There are lots of ways of doing things, and often most of them are right in some given circumstances. So whenever you think "you should never do that", think hard about it. There may be a time, and not just will that lead you to another pattern - it will also help you understand your primary pattern better.
Many people worry about examples in patterns, and code examples in particular. After all patterns are about deep similarities in solutions that look different every time we use them. There is a genuine reason to be worried that some readers will take the example as the pattern, thinking about patterns as glorified macros.
In my view there are many people that understand things much better by example. When given an example, they can then start abstracting for the general principle. That's certainly the way I seem to work. So I'd rather give an example, and run the risk of lack of abstraction, than avoid the example and lose readers completely in the abstraction.
If you are very concerned about a particular interpretation of a pattern, then a useful approach is to use multiple examples. The different examples of the same pattern can help illustrate common threads. Different examples can be different approaches using the same platform, or using different platforms.
Of course some people won't look at code examples, since any code example contains a lot of details. As a result I try to make my code examples skippable - by which I mean that I write the pattern so it's understandable without the code example. Code examples are then a bonus.
When writing a code example there's a tension in how complicated to make it. If I make it too simple, then people may dismiss it as not realistic, but if I make it too complex then people have to understand a bunch of stuff that's nothing to do with the pattern in order to understand the pattern. Complicated enough and we get to what Micheal Feathers calls the MEGO point ("My Eyes Glaze Over") and I've lost them completely. I prefer to err on the side of being too simple. If I get the simple stuff clear then I (or others) can add more complicated stuff with interactions between patterns later. I'd rather people understand a little than fail to understand a lot. This desire is reinforced by the chunking that patterns are about - a reader should only have to read that one pattern to understand the pattern.
Common Pattern Forms
Every author tends to make his own particular pattern form, but certain pattern forms have become more well-known. These are often used exactly by new authors, or at least as starting points.
Many people see Christopher Alexander's A Pattern Language (APL) as an important influence in the patterns world. Alexander write about building architecture and was a big influence behind some of the early advocates of software patterns. He wrote his patterns book in a particular form which is known in the software patterns world as Alexandrian form. As well as the patterns in his book, you can also find good examples of this form in Domain-Driven Design. On the web a good example is Josh Kerievsky's Knowledge Hydrant patterns.
Like many standard forms, we actually see a fair amount of variation of the Alexandrian form in practice. I'll describe it by quoting the description of the form from APL
For convenience and clarity, each pattern has the same format. First, there is a picture, which shows an archetypal example of that pattern. Second, after the picture, each pattern has an introductory paragraph, which sets the context for the pattern, by explaining how it helps to complete certain larger patterns. Then there are three diamonds to mark the beginning of the problem. After the diamonds there is a headline, in bold type. This headline gives the essence of the problem in one or two sentences. After the headline comes the body of the problem. This is the longest section. It describes the empirical background of the pattern, the evidence for its validity, the range of different ways the pattern can be manifested in a building, and so on. Then, again in bold type, like the headline, is the solution - the heart of the pattern - which describes the field of physical and social relationships which are required to solve the stated problem, in the stated context. This solution is always stated in the form of an instruction - so that you know exactly what you need to do, to build the pattern. Then, after the solution, there is a diagram, with labels to indicate its main components.
After the diagram, another three diamonds, to show that the main body of the pattern is finished. And finally, after the diamonds there is a paragraph which ties the pattern to all those smaller patterns in the language, which are needed to complete this pattern, to embellish it, to fill it out.
The patterns in APL average at half a dozen pages each.
The Alexandrian form is a very narrative form, with relatively few headings. As it a result it tends to flow better then most alternatives when you read it. The bolded summary sentences of the problem and the solution stand out well, and allow you to skip through a large body of patterns very quickly.
In software patterns that use the Alexandrian form, a common variation (used by Big Ball of Mud) is to divide up the main section, the body of the problem, into two parts. The first part, after the problem headline, expands on the problem and the issues around it. The second part is moved to after the solution summary, and describes the details of the solution.
I heard Richard Gabriel criticize this, on the grounds that it forces you to duplicate much of the discussion of alternatives and trade-offs. I didn't think too much about this before, but I find I agree with him. Cutting up the body breaks the flow of the pattern and makes it more choppy as you worry about whether issues should be discussed in the problem half or the solution half.
Most patterns books organize the patterns as relatively self-standing sections, like a reference book. Evans embeds the patterns into the flow of a general narrative book. The Alexandrian form helps him do this, as the flow of patterns is more narrative than the more structured pattern forms.
The GOF form is the form used for the seminal Gang of Four book, that really launched patterns into the software world. It's a very structured form, breaking up the pattern into many headings: Intent, Motivation, Applicability, Structure, Participants, Collaborations, Consequences, Implementation, Sample Code, Known Uses, and Related Patterns. The GOF pattern are quite large, a dozen pages each.
The Portland form got its name from the fact that several people, from Portland Oregon, at the first patterns conference all used a similar form. A good on-line example is [cunningham-checks].
Portland form is entirely textual and very short, usually less than a page per pattern. A couple of paragraphs describe the problem, then there is the word "therefore" emphasized typographically, followed by a couple of paragraphs that describe the solution.
So called, because it's most identified with Jim Coplien, I've also heard it referred to as Canonical Form, although I'm not sure which canon such people are referring to. A good on-line example is [coplien-fault-patterns]
I see this form varied a good deal. The key elements are headed sections for Problem, Context, Forces, and Solution. Most authors will add a few extra sections too. Each section is a few paragraphs, with the forces section commonly a list of bullet points. Patterns in this form are usually fairly short - a couple of pages.
This form gets its name from the POSA book. Similarly to GOF it is a very structured and quite large form, although their headings are different: summary, example, context, problem, solution, structure, dynamics, implementation, example resolved, variants, known uses, consequences, and see also. The patterns are usually just over dozen pages in length. An important part of this form is that the patterns are preceded by a narrative chapter that summarizes the patterns and describes the overall topic.
P of EAA Form
I'm being cheeky to call this a standard form, since nobody other than me uses it. But I've been writing patterns for a long time, experimenting with various styles, and this is the one I've come to prefer. It's fairly narrative, with a few sections: how it works, when to use it, and one or more examples. The length averages at around eight pages, but it varies from one page to well over a dozen. This is a recent example.
Choosing Your Pattern Form
As you can see by just looking at the list of common forms, there is a lot of different ways you can write your patterns, in fact there's much more than that. I only mentioned those that are commonly referred to - every book typically uses its own and many papers show many further variations. So my primary piece of advice is to remember that your pattern form is a personal choice. Different forms work for different authors, because different writing styles work with different personalities. The most important thing is to find a form that works with your style of writing, and for the ideas that you want to convey.
A good first step is to begin by reading. Read a lot of different pattern books and papers. Concentrate on the content, but ask yourself which forms seem the most comfortable to you. To really appreciate this you need to read them both in a start-to-end style, but also by looking up and skipping through them. You probably have already done this a lot in the course of your other work - which patterns did you find work best for you?
Once you've got an idea of what form (or forms) you like, start writing. Try to experiment with several different pattern forms. A useful exercise is to write the same pattern in a few different forms to see which ones seem to work best for you. Get some people to review them and tell you which forms seem to read best for them. Don't be afraid to experiment here, it took me many years before I found a pattern form that worked for me.
Once you've picked a basic form, don't let the form force its way too much over the content. I notice this problem particularly with the very structured forms like GOF and POSA - people feel that they have to put something in to every heading for every pattern. But not every pattern requires the same treatment. You'll find some elements that you want in every pattern, but many elements are optional. It's better to leave something out than put in a weak placeholder.
A big question is whether you prefer a more narrative style, or a structured style with many headings. Often when people start out they like the headings because it directs them on how to write. I tend to prefer a more narrative style, because it tends to lead to writing that flows better.
Patterns vary considerably in their size. Portland form often gets a pattern over and done with in a few paragraphs, POSA can go on for a couple of dozen pages. Your choice here depends a great deal on how much detail you want to go into. If you are going to explore implementation issues and provide sample code, you are going to inevitably have longer patterns. In this case, more structure is often more useful, although I use just a few headings even for long patterns.
When you sit down and write patterns, many issues come up for lots of pattern authors. There's not necessarily a right answer to these questions, but at least I can give my perspective on them.
Arranging Patterns into a Structure
A common problem that people run into is how to structure the patterns they are writing. Patterns encourage chunking, and it's easy to concentrate on the chunks. But how do you fit these chunks into something meaningful? I've seen many people struggle with an overall structure to a collection of patterns.
My biggest advice here is "don't worry about it - I don't". I prefer to concentrate on the individual patterns, describing the interesting solutions that I come across. Once I begin to get a set of patterns together I then think about how they should be structured and look for obvious gaps that need more patterns that cover them.
In particular, you don't have to spend a lot of time trying to get an overall structure right before you dig into the patterns. I find I don't really understand the patterns until I dig into the details of describing them.
Remember that in the end it's more valuable to have a bunch of good patterns, poorly organized than it to have a really good structure with weak patterns underneath them.
Patterns and Pattern Languages
I've often been troubled by a lot of the mystique that people summon up around patterns. One of the common areas to kick up this kind of dust is in the issue of patterns and pattern languages. This is often accompanied by 'this isn't a pattern language, it's merely a catalog of patterns'.
The idea behind a pattern language comes again from Alexander. The idea is that you have a body of patterns with a structure that leads you from pattern to pattern. You begin with (usually) some very strategic patterns, each pattern leads you to a point where you have to decide to apply other patterns. A pattern language has a flow that connects the various patterns.
If a pattern language comes easily to you, then well and good - but I don't think a book that's just a loose collection of patterns is a bad thing. Certainly none of my books have been pattern languages, neither is GOF. Pattern languages are also very hard to write - and I've seen people get stuck trying to put them together. Remember that the value of the patterns is the usefulness of what they say, in this sense I look at pattern languages as a structuring mechanism - and the same comments apply to what I said above.
(Knowledge Hydrant is a good on-line example of a pattern language.)
Granularity of Patterns
One of the biggest issues I worry about is how conceptually big to make my patterns. This isn't about how many pages it takes to write them up, but how much conceptual ground is covered by a pattern.
As you start delving into patterns you begin to realize that you often have choices between turning two related concepts into separate patterns, or combining them as variations of a single pattern. A good example of this from GOF is the Proxy pattern, which describes four variations (remote proxy, virtual proxy, protection proxy, and smart pointer). You could write these as four separate patterns, or as one pattern with four variations, or as one summary pattern with four further patterns for each variation.
There's no easy answer to this one, or at least if there is I'd love to know what it is. Deciding where to place the boundaries between patterns is one of the hardest problems I wrestle with.
One thing I do assert is that if you do split them, don't try to have an overall pattern too. So when I worked on patterns for mapping inheritance in object-relational patterns I picked different patterns for Single Table Inheritance, Class Table Inheritance, and Concrete Table Inheritance. I didn't try have an overall "inheritance mapping" pattern to tie them together. This does mean there's some duplication is the 'when' section as I have to discuss the trade-offs between the three in each case. I can live with a some of that duplication (just don't cut and paste the text, write it differently each time). There is a tying together in the narrative - which is part of the purpose of the narrative in my style.
A common question that comes up in reviews of patterns is where a pattern that written for one domain seems to also make sense in others. Alice writes a pattern about database interaction, Bob says that similar advice applies to network communications too and suggests making the pattern more general.
In general, I resist such generalizations. The key question is the author's experience. If Alice knows about databases but not about network programming, then a pattern she writes should describe the database situation. The reader may then consider it applicable to their area of expertise, but it's up to reader to decide if it's applicable. Such a reader is in a better situation to make that judgement than the writer.
Tasks rather than Tools
One of the great plagues of software writing is a focus on tools rather than tasks. Tool oriented books say - “here is a toolbox and I’ll explain how you use each tool”. Task oriented books say “here’s a bunch of tasks you need to do, here’s how to do them (showing you the tools in the process)”. Tool orientation is easier to write, especially for software manuals, because it’s easy to look at a framework (say) and identify the tools.
But task orientation is much better. People don’t approach a book saying “how do I use this widget”, instead they come needing to do some task and trying to find out how to carry that out. With a tools book they spend time looking at likely tools to see if they can help; this is all well and good, but it’s much better if they can find their task right there instead. This is why recipe oriented books are so handy - they focus on tasks.
Patterns always fall into the danger of being tool oriented - after all what we are doing with patterns is trying to identify conceptual tools. It's thus very easy to end up making a book into a tool-oriented guide to newly named tools.
This is the importance of the problem part to patterns. Although in the end we are identifying tools, we can mitigate the dangers of tool-orientation by putting a lot of thought into the problem each pattern (tool) solves. We are free to shape the pattern boundaries how we choose, indeed this freedom is what makes pattern writing so hard. Trying to make the work as task-oriented as we can helps draw those boundaries in a useful way.
Nothing new here
A common complaint about patterns books is that they have nothing new to tell experienced developers. Not just is this true, but it's the whole point of patterns.
Patterns are there to capture knowledge from the field, not to present original ideas. As a result it's inevitable that patterns books aren't going to add stunning new ideas to those who have been working in a field for a while. But even so I think there's an important role for patterns books even to those who don't need to learn the ideas. This role is to help the experienced people communicate their experience to those around them who are less experienced. Few teams consist of entirely of seasoned developers. One of the most important things an experienced leader can do is pass on her skills.
So if you're evaluating some patterns in a field you're an expert in, don't expect to learn new things. Instead evaluate them on how they would help you to communicate your knowledge to others. Try them out by using them yourself and seeing if they help people grasp important concepts.
This is why patterns books should also age well. Many of the fundamentals of software design don't change very rapidly, even though our technologies do. So don't be concerned too much if a pattern book is old.
Despite the recursiveness that borders on cutesyness, the patterns of Meszaros and Doble form a worthwhile set of further advice on pattern-writing.
03 August 2020: Added intro paragraphs, fixed some links.
01 August 2006: Brushed up for initial publication.
April 2003: Made an initial draft but never published it.