This Lambda The Ultimate discussion on Benjamin C. Pierce's (Types And Programming Languages) talk Types Considered Harmful is also of interest - the gist being that contracts are useful because sometimes static types are just too damn hard - http://lambda-the-ultimate.org/node/2828
EDIT: Might as well link to Brendan Eich's talk Proxies Are Awesome!, http://www.slideshare.net/BrendanEich/metaprog-5303821. I haven't really been following the Proxy thing and they seem quite useful. It's nice to see them applied so powerfully here.
I ♥ where it's going! "Nhaa, I won't use a statically typed language.. I prefer javascript with coffeescript syntax. But, I'll use contracts and tests to make it robust." Hands down, it's great and I'll definitely use it.
guardArgs = (def, args) ->
i = 0
for arg, type of def
throw new Error("bad type of #{arg}") unless typeof args[i] == type
i++
typeEnsure = (def, f) -> ->
guardArgs(def, arguments)
f.apply @, arguments
myFunc = typeEnsure {name: 'string', i: 'number'}, (name, i) ->
console.log name, i
myFunc("foo", 1) # ok
myFunc("foo", "x") # Error: bad type of i
Neat. I built runtime argument and return value type-checking into Objective-J awhile back (not enabled by default due to performance overhead). It was really easy since we already declare Objective-J argument and return value types, and we can easily intercept every Objective-J message invocation.
How does it integrate with the `coffee` executable? Does coffeescript have a compiler plugin system, or are you patching the compiler? Will this cause me problems when I try to update my coffeescript compiler?
It's a fork of CoffeeScript so it replaces your "coffee" executable. If you don't want to overwrite your normal CoffeeScript compiler just run the contracts.coffee "coffee" executable from its own directory.
No ambiguity; if you put brackets around it you get a tuple (really, an array of fixed elements), and without brackets, the comma is part of the function signature. If you wanted to express your interpretation #1 you'd write:
This is great, it's like Google's compiler library[0] but with run-time checking! Coffee script is still very hard to debug, this should help avoid about 80% of the mistakes I normally make when coding.
ok, figured out that it takes care of the existing prototype operator (::) in lexing [1],[2]. It seems little annoying but guess I'll get the hang of it.
very cool project!
Out of curiosity, what is the overhead of a program with-contracts vs one without them? It kind of looks like a JIT should be able to optimize away most of the guards.
Also, I believe other DBC platforms had options to disable run time enforcing of contracts, is this possible somehow with CFCS ?
I haven't done any benchmarks yet so not sure. In addition to the guards I think the use of Proxies will cause some slowdown. But you're right the JIT should be able to optimize most of those guards away (maybe even someday use the guards to help in their optimizing/tracing...just speculating here, I'm not a compiler/JIT guy).
If you compile with "coffee -c" (compile, no contracts) instead of "coffee -cC" (compile, with contracts) you get pure JavaScript, no contracts and no runtime checks. Exactly what vanilla CoffeeScript would have given you.
I believe jashkenas' example defines both a pre-condition and a post-condition, specifying that f accepts only odd numbers and returns only even numbers ('!' here is the operator for new contract, rather than logical negation). If f is called with anything other than an odd number, an error will be thrown right away, before the body of the function is run.
Those '!'s are really confusing. From the example in the submission:
isEven = (x) -> x % 2 is 0
isOdd = (x) -> x % 2 isnt 0
addEvens :: (!isEven) -> !isOdd
addEvens = (x) -> x + 1
I am guessing that both '!'s define a new contract - and also guessing that coffeescript does not actually have a '!' operator for logical negation. So that addEvens simply takes an even number and returns an odd number.
I find it confusing that the different syntax highlighting of the two '!' suggests (wrongly) that they have different meanings. Also, logical negation is something often used with isOdd/isEven functions.
CoffeeScript does allow ! for negation, but contracts have their own syntax and semantics separate from the rest of the language. It's a little confusing initially, but I suppose there aren't a lot of good operators left. There are a lot of similar convenient inconsistencies that people hardly notice once they sink in:
• Indentation shifts mean different things depending on context (could mean we're doing an object literal, could mean we're defining a function, could mean we're entering a loop, etc.)
• Looping through arrays uses "in" while looping through everything else uses "of"
• Array and object literal syntax might create arrays and objects or define variables depending on which side of the equals sign you're reading
How is this different than creating an OddNumber and EvenNumber type that check their input in the constructor at run-time? That is the point of a type system.
Scala can make this really transparent and concise using an abstract class to implement most of the plumbing and implicit type conversion to automatically convert the contract enforcement types to and from the native types.
case class EvenNumber(val x: Int) extends Contract[Int] {
require(x % 2 != 0)
}
That while this looks like a good extension to CoffeeScript, it's unfortunate that Haskell's type syntax was appropriated when the relationship between contracts and types is tenuous at best.
There are good uses for static type checkers and design by contract, but neither one is a good substitute for the other.
I'm not sure that the relationship between types and contracts is all that tenuous.
They both work to enforce program invariants and there's a large overlap in the invariants they can express. Usually contracts are checked dynamically (as in contracts.coffee) but this is not a hard requirement (for example .net's code contracts can be statically checked [1]).
It's a grammar hack, but IMO an intuitive one: if you have whitespace around the ::, it's a contract annotation; otherwise it's the prototype operator.
To quote Jeremy Ashkenas's JSConf talk on forking CoffeeScript: "when in doubt, cheat." ;-P
I feel a little bit crazy because no one else has brought this up yet, but what's the point of declaring types if your errors don't get caught at compile time? If they're going to throw errors at runtime, how is that an improvement over not checking the types?
Well, CoffeeScript/JavaScript doesn't have a static type system. And to add one would be...hard.
But we would still like to express and enforce some invariants about our code. And we'd like to do it in a more elegant way than constantly writing "if(argument === ...) then ... else ..."
Contracts also allow us to check invariants that might not result in a run-time exception but could lead to the code just being subtly wrong.
> Well, CoffeeScript/JavaScript doesn't have a static type system. And to add one would be...hard.
Indeed it would be. That said, CoffeeScript is a compiler of sorts.
> And we'd like to do it in a more elegant way than constantly writing "if(argument === ...) then ... else ..."
This doesn't really seem to solve that problem. It just throws consistent errors if the types don't match. Your code would likely throw its own error if you didn't check the type manually. In my experience, native js errors are as easy or easier to track down that throwing errors yourself.
> Contracts also allow us to check invariants that might not result in a run-time exception but could lead to the code just being subtly wrong.
Ah, ok. That makes more sense. They're like python decorators, only single purpose.
> Indeed it would be. That said, CoffeeScript is a compiler of sorts.
The fact that we have a compiler doesn't give us all that much. Adding types to a dynamic language (where things can be monkey patched, evaled, and just generally mutated willy-nilly) is a hard problem (not impossible but definitely research [1 and 2 to start, but much more]).
>> And we'd like to do it in a more elegant way than constantly writing "if(argument === ...) then ... else ..."
> This doesn't really seem to solve that problem.
It solves the problem of elegance :)
> It just throws consistent errors if the types don't match. Your code would likely throw its own error if you didn't check the type manually. In my experience, native js errors are as easy or easier to track down that throwing errors yourself.
Key word there is likely. Like I mentioned, there are some errors that don't result in a run-time error and we might like to check for those too. As a silly example, consider a binary search tree [3] that you want to keep balanced. If you get the balance wrong, there's no immediate fault (no TypeError/NPE).
> Ah, ok. That makes more sense. They're like python decorators, only single purpose.
I guess you could say that. If we had an expressive enough decorator-like system for CoffeeScript we could probably implement contracts with it. CS doesn't have one of course so path of least resistance is to add the contract language extension directly.
This Lambda The Ultimate discussion on Benjamin C. Pierce's (Types And Programming Languages) talk Types Considered Harmful is also of interest - the gist being that contracts are useful because sometimes static types are just too damn hard - http://lambda-the-ultimate.org/node/2828
EDIT: Might as well link to Brendan Eich's talk Proxies Are Awesome!, http://www.slideshare.net/BrendanEich/metaprog-5303821. I haven't really been following the Proxy thing and they seem quite useful. It's nice to see them applied so powerfully here.