The Universe of Discourse
           
Tue, 03 Oct 2006

Ralph Johnson on design patterns
Last month I wrote an article about design patterns which attracted a lot of favorable attention in blog world. I started by paraphrasing Peter Norvig's observation that:

"Patterns" that are used recurringly in one language may be invisible or trivial in a different language.

and ended by concluding:

Patterns are signs of weakness in programming languages.

When we identify and document one, that should not be the end of the story. Rather, we should have the long-term goal of trying to understand how to improve the language so that the pattern becomes invisible or unnecessary.

Ralph Johnson, one of the four authors of the famous book Design Patterns, took note of my article and responded. I found Johnson's response really interesting, and curious in a number of ways. I think everyone who was interested in my article should read his too.

[ Addendum 20070127: The link above to Ralph Johnson's response is correct, but your client will be rejected if you are referred from here. To see his blog page, visit the page without clicking on the link. ]

Johnson raises several points. First there is a meta-issue to deal with. Johnson says:

He clearly thinks that what he says is surprising. And other people think it is surprising, too. That is surprising to me.
I did think that what I had to say was interesting and worth saying, of course, or I would not have said it. And I was not surprised to find that other people agreed with me.

One thing that I did find surprising is the uniformity of other people's surprise and interest. There were dozens of blog posts and comments in the following two weeks, all pretty much saying what a great article I had written and how right I was. I tracked the responses as carefully as I could, and I did not see any articles that called me a dumbass; I did not see any except for Johnson's that suggested that what I was saying was unsurprising.

We can't conclude from this that I am right, of course; people agree with all sorts of stupid crap. But we can conclude that that what I said was surprising and interesting, since people were surprised and interested by it, even people who already have some knowledge of this topic. Johnson is right to be surprised by this, because he thought this was obvious and well-known, and that it was clearly laid out in his book, and he was mistaken. Many or most of the readers of his book have completely missed this point. I didn't miss it, but I didn't get it from the book, either.

Johnson and his three co-authors wrote this book, Design Patterns, which has had a huge influence on the way that programming is practiced. I think a lot of that influence has been malign. Any practice can be corrupted, of course, by being reduced to its formal aspects and applied in a rote fashion. (There's a really superb discussion of this in A. Ya. Khinchin's essay On the Teaching of Mathematics, and a shorter discussion in Polya's How to Solve It, in the section on "Pedantry and Mastery".) That will happen to any successful movement, and the Gang of Four can't take all the blame for that.

But if they really intended that everyone should understand that each design pattern is a demonstration of a weakness in its target language, then they blew it, because it appears that hardly anyone understood that.

Let's pause for a moment to imagine an alternate universe in which the subtitle of the Design Patterns book was not "Elements of Reusable Object-Oriented Software" but "Solutions for Recurring Problems in Object-Oriented Languages". And let's imagine that in each section, after "Pattern name", "Intent", "Motivation", "Applicability", and so forth, there was another subsection titled "Prophylaxis" that went something like this: "The need for the Iterator pattern in C++ appears to be due partly to its inflexible type system and partly to its lack of abstract iteration structures. The iterator pattern is unnecessary in the Python language, which avoids these defects as follows: ... at the expense of ... . In Common Lisp, on the other hand, ... (etc.)".

I would have liked to have seen that universe, but I suppose it's too late now. Oh well.

Anyway, moving on from meta-issues to the issues themselves, Johnson continues:

At the very end, he says that patterns are signs of weakness in programming languages. This is wrong.
This is interesting, and I was going to address it later, but I now think that it's the first evidence of a conceptual mistake that Johnson has made that underlies his entire response to my article, so I'll take it up now.

At the very end of his response, Johnson says:

No matter how complicated your language will be, there will always be things that are not in the language. These things will have to be patterns. So, we can eliminate one set of patterns by moving them into the language, but then we'll just have to focus on other patterns. We don't know what patterns will be important 50 years from now, but it is a safe bet that programmers will still be using patterns of some sort.
Here we are in complete agreement. So, to echo Johnson, I was surprised that he would think this was surprising. But how can we be in complete agrement if what I said was "wrong"? There must be a misunderstanding somewhere.

I think I know where it is. When I said "[Design] Patterns are signs of weakness in programming languages," what I meant was something like "Each design pattern is a sign of a weakness in the programming language to which it applies." But it seems that Johnson thinks that I meant that the very existence of design patterns, at all, is a sign of weakness in all programming languages everywhere.

If I thought that the existence of design patterns, at all, was a sign that current programming languages are defective, as a group, I would see an endpoint to programming language development: someday, we would have a perfect überlanguage in which it would be unnecessary to use patterns because all possible patterns would have been built in already.

I think Johnson thinks this was my point. In the passage quoted above, I think he is addressing the idea of the überlanguage that incorporates all patterns everywhere at all levels of abstraction. And similarly:

Some people like languages with a lot of features. . . . I prefer simple languages.

And again:

No matter how complicated your language will be, there will always be things that are not in the language.

But no, I don't imagine that someday we will have the ultimate language, into which every conceivable pattern has been absorbed. So a lot of what Johnson has to say is only knocking down a straw man.

What I imagine is that when pattern P applies to language L, then, to the extent that some programmer on some project finds themselves needing to use P in their project, the use of P indicates a deficiency in language L for that project.

The absence of a convenient and simple way to do P in language L is not always a problem. You might do a project in language L that does not require the use of pattern P. Then the problem does not manifest, and, whatever L's deficiencies might be for other projects, it is not deficient in that way for your project.

This should not be difficult for anyone to understand. Perl might be a very nice language for writing a program to compile a bioinformatic data file into a more reasonable form; it might be a terrible language for writing a real-time missile guidance system. Its deficiencies operate in the missile guidance project in a way that they may not in the data munging project.

But to the extent that some deficiency does come up in your project, it is a problem, because you are implementing the same design over and over, the same arrangement of objects and classes, to accomplish the same purpose. If the language provided more support for solving this recurring design problem, you wouldn't need to use a "pattern". Consider again the example of the "subroutine" pattern in assembly language: don't you have anything better to do than redesign and re-implement the process of saving the register values in a stack frame, over and over? Well, yes, you do. And that is why you use a language that has that built in. Consider again the example of the "object-oriented class" pattern in C: don't you have anything better to do than redesign and re-implement object-oriented method dispatch with inheritance, over and over? Yes, you do. And that is why you use a language that has that built in, if that is what you need.

By Gamma, Helm, Johnson, and Vlissides' own definition, the problems solved by patterns are recurring problems, and programmers must address them recurringly.

If these problems recurred in every language, we might conclude that they were endemic to programming itself. We might not, but it's hard to say, since if there are any such problems, they have not yet been brought to my attention. Every pattern discovered so far seems to be specific to only a small subset of the world's languages.

So it seems a small step to conclude that these recurring, language-specific problems are actually problems with the languages themselves. No problem is a problem in every language, but rather each problem is a red arrow, pointing at a design flaw in the language in which it appears.

Johnson continues:

Patterns might be a sign of weakness, but they might be a sign of simplicity. . . .
I think this argument fails, in light of the examples I brought up in my original article. The argument is loaded by the use of the word "simplicity". As Einstein said, things should be as simple as possible, but no simpler. In assembly language, "subroutine call" is a pattern. Does Johnson or anyone seriously think that C++ or Smalltalk or Common Lisp or Java would be improved by having the "subroutine call" pattern omitted? The languages might be "simpler", but would they be better?

The alternative, remember, is to require the programmer to use a "pattern": to make them consult a manual of "patterns" to implement a "general arrangement of objects and classes" to solve the subroutine-call problem every time it comes up.

I guess you could interpret that as a sign of "simplicity", but it's the wrong kind of simplicity. Language designers have a hard problem to solve. If they don't put enough stuff into the language, it'll be too hard to use. But if they put in too much stuff, it'll be confusing and hard to program, like C++. One reason it's hard to be a language designer is that it's hard to know what to put in and what to leave out. There is an extremely complex tradeoff between simplicity and functionality.

But in the case of "patterns", it's much easier to understand the tradeoff. A pattern, remember, is a general method for solving "a recurring design problem". Patterns might be a sign of "simplicity", but if so, they are a sign of simplicity in the wrong place, a place where the language needs to be less simple and more featureful. Because patterns are solutions to recurring design problems.

If you're a language designer, and a "pattern" comes to your attention, then you have a great opportunity. The programmers using your language have a recurring problem. They have to implement the same solution to it, over and over. Clearly, this is a good place to try to expend some design effort; perhaps you can trade off a little simplicity for some functionality and fix the language so that the problem is a problem no longer.

Getting rid of one recurring design problem might create new ones. But if the new problems are operating at a higher level of abstraction, you may have a win. Getting rid of the need for the "subroutine call" pattern in assembly language opened up all sorts of new problems: when and how do I do recursion? When and how do I do coroutines?

Getting rid of the "object-oriented class" pattern in C created a need for higher-level patterns, including the ones described in the Design Patterns book. When people didn't have to worry about implementing inheritance themselves, a lot of their attention was freed up, and they could notice patterns like Façade.

As Alfred North Whitehead says, civilization advances by extending the number of important operations which we can perform without thinking about them. The Design Patterns approach seems to be to identify the important operations and then to think about them over and over and over and over and over.

Or so it seems to me. Johnson's next paragraph makes me wonder if I've completely missed his point, because it seems completely senseless to me:

There is a trade-off between putting something in your programming language and making it be a convention, or perhaps putting it in the library. Smalltalk makes "constructor" be a convention. Arithmetic is in the library, not in the language. Control structures and exception handling are from the library, not in the language.
Huh? Why does "library" matter? Unless I have missed something essential, whether something is in the "language" or the "library" is entirely an implementation matter, to be left to the discretion of the compiler writer. Is printf part of the C language, or its library? The library, everyone knows that. Oh, well, except that its behavior is completely standardized by the language standard, and it is completely permissible for the compiler writer to implement printf by putting a special case into the compiler that is enabled when the compiler happens to see the directive #include <stdio.h>. There is absolutely no requirement that printf be loaded from a separate file or anything like that.

Or consider Perl's dbmopen function. Prior to version 5.000, it was part of the "language", in some sense; in 5.000 and later, it became part of the "library". But what's the difference, really? I can't find any.

Is Johnson talking about some syntactic or semantic difference here? Maybe if I knew more about Smalltalk, I would understand his point. As it is, it seems completely daft, which I interpret to mean that there's something that went completely over my head.

Well, the whole article leaves me wondering if maybe I missed his point, because Johnson is presumably a smart guy, but his argument about the built-in features vs. libraries makes no sense to me, his argument about simplicity seems so clearly and obviously dismantled by his own definition of patterns, and his apparent attack on a straw man seems so obviously erroneous.

But I can take some consolation in the thought that if I did miss his point, I'm not the only one, because the one thing I can be sure of in all of this is that a lot of other people have been missing his point for years.

Johnson says at the beginning that he "wasn't sure whether to be happy or unhappy". If I had written a book as successful and widely read as Design Patterns and then I found out that everyone had completely misunderstood it, I think I would be unhappy. But perhaps that's just my own grumpy personality.

[ Addendum 20080303: Miles Gould wrote a pleasant and insightful article on Johnson's point about libraries vs. language features. As I surmised, there was indeed a valuable point that went over my head. I said I couldn't find any difference between "language" and "library", but, as M. Gould explains, there is an important difference that I did not appreciate in this context. ]


[Other articles in category /prog] permanent link