Dynamic Types FTW

There's a belief among some programmers, particularly the OO classical inheritance folks, that static typing is a bulwark of security against all kinds of disasters that may happen in your code.  They feel that the compiler will ensure their programs work correctly, by generating errors and warnings and refusing to compile their program until all the mistakes are fixed.

Some developers mistakenly speak in terms of "strong typing" and "weak typing". Put that way, who wouldn't prefer "strong" over "weak" typing? In reality the terms "strong typing" and "weak typing" are undefined in the language of computer science. They're misnomers that people sometimes use to talk about strict (explicit) type definition with static type checking, versus dynamic (runtime) type checking with implicit type conversion.

It's unfortunate that this line of reasoning causes some people to avoid languages that offer dynamic types. Dynamic types can be very useful, and static typing actually offers very little in the way of ensuring program correctness.  I finally sat down and watched Eric Elliott's presentation "Static Types are Overrated: The Dynamic Duo - Loose Types and Object Extension" and was treated to a bang-on description why static types aren't all they're cracked up to be.  Eric does a good job of explaining and clarifying things about JavaScript and programming in general that I intuitively believed but had a difficult time putting into words or backing up with solid examples.

I borrowed a bunch of points from Eric's talk when I spoke at a recent Ottawa JavaScript meetup. I was presenting on Flow, a static type checker for JavaScript, but I wanted people to see it as something that could be added to their regular code linting process rather than something to enforce a statically typed programming style. Here are some of them.
  • It's a myth that dynamic loosely typed languages like JavaScript aren't suitable for building large, complex applications.  There are dozens of examples from Facebook to 37signals, Dow Jones, Adobe, Flickr, LinkedIn, Walmart, etc).
  • Type correctness does not guarantee program correctness. This is one of the biggest cargo-cult myths out there: thinking that static, strict type checking will save you. It won't.
  • When languages lack dynamic types, people fake it with ugly hacks - void pointers, variadic functions, abusing array accessors, type-casting of all sorts, generics and ugly template classes, to name a few. I don't know any C or Java programmers who haven't used and abused at least a few of these techniques.
  • "Any sufficiently advanced C / Fortran [Java, etc] program contains an ad hoc, informally specified, bug-ridden, slow implementation of half of common Lisp" - Greenspun's Tenth Rule. (There are no rules 1 through 9, by the way. Greenspun's Rules of Programming start and end with number 10.)
  • Older statically typed languages are now introducing dynamic types because they really are useful. Objective-C added them a long time ago; more recently C++11 has added dynamic types, Java has "generics", and libraries like cppscript and boost provided ways to do dynamic types in earlier versions of C++.
  • Functional programming benefits from functions that can operate on any type which implements its requirements (for example, arguments have a valueOf method). Otherwise known as duck typing. This permits things like map, filter, forEach etc. to be used generically.
  • Correctness can only really be assured through proper unit and integration testing. 
  • Code that is well organized in small, simple modules, linted, unit tested, peer reviewed, and integration tested, is very unlikely to contain type errors. 

The last thing I would add is that while static types are over-rated, static code analysis tools like JSLint, Tern.js and Flow are mostly underrated and underutilized. If you use something like Flow or Tern.js to provide hints and insights into potential type mismatches, the chances of type errors become essentially zero. Ideally they should be as up-front as possible, i.e. integrated directly into your code editor. I wrote a JSLint plugin for SublimeText and recently wrote a Flow plugin as well. They provide immediate static code analysis warnings as I edit my code, so I can fix problems before I commit anything to the source code repository.

Comments

Wolfie Wolf said…
You're 100% correct. And all the die hard C# guys overlook the fact that XAML is a dynamic language. The compiler does next to nothing to validate the code and what's worse is that it compiles and then silently fails creating a troubleshooting nightmare. If M$ wasn't so worried about the "purists" they would exploit the dynamic nature of XAML and focus on runtime error reporting. I've spent far too much time trying to use C# like a dynamic language. Roslyn is actually a step in the wrong direction from the DLR. Oh and Hi from PEI. I'm playing around with dclassify and it's looking good!
Wolfie Wolf said…
You're 100% correct. And all the die hard C# guys overlook the fact that XAML is a dynamic language. The compiler does next to nothing to validate the code and what's worse is that it compiles and then silently fails creating a troubleshooting nightmare. If M$ wasn't so worried about the "purists" they would exploit the dynamic nature of XAML and focus on runtime error reporting. I've spent far too much time trying to use C# like a dynamic language. Roslyn is actually a step in the wrong direction from the DLR. Oh and Hi from PEI. I'm playing around with dclassify and it's looking good!

Popular posts from this blog

Selenium IDE Flow Control - Goto and While Loops

REST API Best Practices 4: Collections, Resources and Identifiers

REST API Best Practices 3: Partial Updates - PATCH vs PUT