We missed the point when we adopted "patterns" in the software world. Instead of an organic whole, we got a bag of tricks.
The commonly accepted definition of a pattern is "a solution to a problem in a context." This is true, but limiting. This definition loses an essential characteristic of patterns: Patterns relate to other patterns.
We talk about the context of a problem. "Context" is a mental shorthand. If we unpack the context it means many things: constraints, capabilities, style, requirements, and so on. We sometimes mislead ourselves by using the fairly fuzzy, abstract term "context" as a mental handle on a whole variety of very concrete issues. Context includes stated constraints like the functional requirements, along with unstated constraints like, "The computation should complete before the heat death of the universe." It includes other forces like, "This program is written in C#, so the solution to this problem should be in the same language or a closely related one." It should not require a supercooled quantum computer, for example.
Where does the context for a small-scale pattern originate?1 Context does not arise ex nihilio. No, the context for a small-scale pattern is created by larger patterns. Large grained patterns create the fabric of forces that we call the context for smaller patterns. In turn, smaller patterns fit into this fabric and, by their existence, they change it. Thus, the small scale patterns create feedback that can either resolve or exacerbate tensions inherent in the larger patterns.
Solutions that respect their context fit better with the rest of the organic whole. It would be strange to be reading some Java code, built into layered architecture with a relational database for storage, then suddenly find one component that has its own LISP interpreter and some functional code. With all respect to "polyglot programming", there’d better be a strong motivation for such an odd inclusion. It would be a discontinuity… in other words, it doesn’t fit the context I described. That context—the layered architecture, the OO language, relational database—was created by other parts of the system.
If, on the other hand, the system was built as a blackboard architecture, using LISP as glue code over intelligent agents acting asynchronously, then it wouldn’t be at all odd to find some recursive lambda expressions. In that context, they fit naturally and the Java code would be an oddity.
This interrelation across scale knits patterns together into a pattern language. By and large, what we have today is a growing group of proper nouns. Please don’t get me wrong, the nouns themselves have use. It’s very helpful to say "you want a Null Object there," and be understood. That vocabulary and the compression it provides is really important.
But we shouldn’t mistake a group of nouns for a real pattern language. A language is more than just its nouns. A language also implies ways of connecting statements sensibly. It has idioms and semantics and semiotics.2 In a language, you can have dialog and argumentation. Imagine a dialog in patterns as they exist today:
"Pipes and filters."
"Chain of Responsibility!"
You might be able to make a comedy sketch out of that, but not much more. We cannot construct meaningful dialogs about patterns at all scales.
What we have are fragments of what might become a pattern language. GoF, the PLoPD books, the PoSA books… these are like a few charted territories on an unmapped continent. We don’t yet have the language that would even let us relate these works together, let alone relating them to everything else.
Everything else? Well, yes. By and large, patterns today are an outgrowth of the object-oriented programming community. I contend, however, that "object-oriented" is a pattern! It’s a large-scale pattern that creates really significant context for all the other patterns that can work within it. Solutions that work within the "object-oriented" context make no sense in an actor-oriented context, or a functional context, or a procedural context, and so on. Each of these other large-scale patterns admit different solutions to similar problems: persistence, user interaction, and system integration, to name a few. I can imagine a pattern called "Event Driven" that would work very well with "Object oriented", "Functional", and "Actor Oriented", but somewhat less well with "Procedural programming", and contradict utterly with "Batch Processing". (Though there might be a link between them called "Buffer file" or something like that.)
That’s the piece that we missed. We don’t have a pattern language yet. We’re not even close.
1. By "large" and "small", I don’t mean to imply that patterns simply nest hierarchically. It’s more complex and subtle than that. When we do have a real pattern language, we’ll find that there are medium-grained patterns that work together with several, but not all, of the large ones. Likewise, we’ll find small-scale patterns that make medium sized ones more or less practical. It’s not a decision tree or a heuristic.
2. That’s what keeps, "Fill the idea with blue" from being a meaningful sentence. All the words work, and they’re even the right part of speech, yet the sentence as a whole doesn’t fit together.