This 100x! I don't understand how people can work on a complex codebase without a strong static type system. I've tried it once, and still remember the lack of confidence after every non-trivial refactoring.
I've worked on multiple large Java codebases, and on semi-large Clojure code bases. The Java ones compiled great but were full of bugs obscured by 100s of useless classes with bespoke interfaces. I found the Clojure code bases to be far more tractable because the language leads to better opportunities for abstraction and more concise code with way less incidental complexity. I think spec can add another layer of description where it's useful (and is not required where it's not).
That's anecdotal, and probably more a consequence of the seniority of the engineers who wrote these two code bases than the nature of the type systems.
My point about static type systems is broader and actually backed by science and mathematics. And also, just observing the trend and seeing where the wind blows.
You'll never get it because there isn't any. Static vs dynamic typing is largely a matter of preference despite all the hooting from the recent static typing fad. I truly wish this "apparent truism" would go away, since it commonly turns to people flaming dynamic languages as somehow inherently lesser than those with big fancy type checkers.
Indeed. I’ve googled for this research, and there isn’t much of it - just something that “everybody knows”.
The best, recent thing I found was a study done about 5 years ago or so (I think) by a fellow at UC Davis, and the data were pretty weak. Functional programming, and the requisite memory management, had at least as big an effect as static typing.
At this point some Haskell snob steps in and say “sound typing”... :-)
I guess that's why all those statically typed programs don't have any bugs and don't need any tests.
Being "sound" just means you can prove things are consistent with the assumptions you can encode in it. If there are things you can't encode in the types, or if your assumptions are wrong (misapprehension being one of the main problems in software), or if your requirements change, soundness is not going to save you.
Static types and proofs are valuable. Programs like compilers have fixed inputs and outputs and are excellent places to lean on things you can prove. But most of the programs I've worked on are not like compilers. They run for years, the requirements change, they have to deal with dirty data, talk to other messy systems, etc. And I'm not saying that dynamic types are perfect either.
My point is simply that static types are not a magical end goal of programming. They are a tool with tradeoffs.
> I guess that's why all those statically typed programs don't have any bugs and don't need any tests.
This is both a straw man and a fallacy. Nicely done.
Also something I never claimed.
You are basically saying that since you can't prove a program can be 100% demonstrated to be correct, there's no point in trying to approach this goal, therefore dynamically typed languages are good enough.
With your reasoning, we wouldn't be using safety belts because they can't guarantee you'll survive a car crash 100%.
Sure, static types have trade offs, but they are a clear improvement on all dimensions over dynamic types, which is why we see dynamic languages converging toward static typing and never the other way around.
In ten years from now, we'll look at dynamically typed languages as "It looked cool at the time, but we know better now". A bit like we look at FORTRAN and COBOL today.
> Sure, static types have trade offs, but they are a clear improvement on all dimensions over dynamic types
That has not been my experience and I don't think there is much objective proof of this. Most of the (admittedly not great) studies I've seen show dynamic languages as comparing favorably or better in bug counts for example.
Some of the things we're working on in the next version of spec are is head-on the notion of how to define expressive contracts for functions and allow those to meaningfully evolve in compatible ways over time as program requirements change.
I look at FORTRAN as a great language for its domain, so maybe you're right. :)
In a nutshell, if you can mathematically prove a certain degree of soundness of your code, you have not done anything to show your code does what it is supposed to do.
Basically, you could be doing wasted work to satisfy your compiler, and have made zero progress in solving your problem, because that's not what the type system does.
Dynamical languages focuses on solving problems of human beings. The human beings' requirements are often inherently arbitrary, logically unsound, or even self-contradictory. Shoehorning a type system in there often serves to make solving human's problem harder. Consequently, in practice, most popular programming languages are dynamic. The current craze with types is just a fad, in my opinion.
> Dynamical languages focuses on solving problems of human beings
> Shoehorning a type system in there often serves to make solving human's problem harder.
For the first pass author maybe. Subsequent readings and modifications of the code by other authors are much, much harder without "annotated" code (be that types or schema like spec)
Having worked on many codebases of all sizes, hand written by an ever changing population of developers (contractors, interns, consulting companies, etc.) I don't look at such annotation as the holy grail of developer-to-developer communication (in this respect, I find them quite less powerful than tests), nor as the philosophical stone of coping with changing requirements.
