Go vs. Swift [pdf](github.com) |
Go vs. Swift [pdf](github.com) |
The interface concept of Go makes programming with composition much more flexible and powerful than with the class model. The author skips this Go specific and original interface typing. This provides a multiple inheritance equivalent without all the complications on C++ and that most OO oriented languages forbid because of that complication.
Go is a very original language in this aspect as well as with concurrency. Understanding and mastering these properties goes beyond simple syntax analysis.
To me the most remarkable property of Go is its simplicity. As I explained to a friend who is a strong advocate of D, the difference with other programming language is the same as dealing with a spoken language of 1000 words instead of 10,000 words. It's true that the language with 10,000 words is more expressive and richer. But the effort required to learn, read and write a language of 1000 words is much lower than a with a language of 10000 words. I'm beyond 50 years old, and too me this makes a huge difference. The best way to express it is that with Go programming is fun again. I hope that Go will preserve this simplicity. At the beginning Java was simple too. They later killed it to the point I don't want to deal with Java code anymore.
You can compare OOP with parametric polymorphism, you can compare it with type-classes. Heck, OOP isn't necessarily about subtyping and we could be talking about row polymorphism (e.g. OCaml) which has some really nice properties.
> This provides a multiple inheritance equivalent without all the complications on C++ and that most OO oriented languages forbid because of that complication.
Except that it doesn't solve the fundamental problems with OOP, because it's still essentially OOP with subtyping ... and lots of marketing. So coming from Go you can be excused for thinking that the last 30 years of research have been for nothing.
I'm not sure what fundamental problems you're speaking of, but what's nice about Go viz-a-vis C++ is that classes are fundamentally open. Personally, I think it's a win. By the way, I'm a C++ dev and I love C++ too. I just think Go has really interesting ideas.
Also, `Go is a very original language in this aspect as well as with concurrency` CSP is quite old also.
Dlang, is quite a nice and easy to start language, If I will compare d vs go, I will say that you can see that golang has a lot of money and people behind. And, after I wrote more than a simple library, I tend to be tired of some parts of the language decision, and is not about interface, is just the small stuff. But again, I tend to write code that is more OOP :)
Please excuse me if my reply offend you, that is just my personal opinion after work with golang few years ago
Actually it isn't.
Using object Composition/Aggregation is a very old, and composition as the recommended paradigm dates back at least to COM (that's just my memory, it's probably older). Granted, delegation in COM aggregates was based on conventional interface and not structural typing, and it was implemented using ugly C macros, but dynamic languages made it easier and far more flexible. Kotlin even made it part of the language: https://kotlinlang.org/docs/reference/delegation.html
What Go has, despite all the hype, is not real composition with delegation though, but a crippled form of inheritance. With proper composition, your child objects are named, and you can choose which interfaces you want to delegate to which object. This doesn't just give you better control over which functionality you wish to expose, but also avoids conflict when two member objects implement the same interface. In Go you essentially get the same nasty diamond problem of multiple inheritance, with none of the benefits. Sure, you can disambiguate your method calls, but you could do the same in C++ and Python.
As for Go's approach for concurrency obviously isn't new. The CSP model was implemented by Rob Pike in at least 3 different languages previous to Go (Newsqueak, Alef and Limbo) and theory behinds it dates to research by Tony Hoare in the late 1970s.
I won't argue with you that Go is simple, but it's not revolutionary. As for the fun, I think that really depends on the individual. I know many people who think Go is fun, but for others, like me, it's about as fun as fingernails on chalkboard. There seem to be a strong correlation between people who like functional programming (especially the ML camp, but also the LISP camp).
For me coding in Go is plain drudgery: error handling boilerplate, noisy imperative looping (and in general very low SNR) and almost no ways to create abstractions. Yeah, you can make very useful and down to earth software in this language and it's currently satisfying much of my microservices needs at work. But it isn't what I would call fun.
Oh but you can avoid that, it's part of the language, see this link:
https://golang.org/doc/effective_go.html#embedding
Typically, you can have something like this:
type lockedReader struct {
io.Reader
sync.Mutex
}
lr := lockerReader{someReader, sync.Mutex{}}
lr.Lock()
lr.Read(...)
lr.Unlock()
Nailed it.
I beg to disagree in the strongest possible terms. I found programming with Go to be the opposite of fun. It's a highly opinionated language which gives people poor libraries, poor tooling, and the worst of many possible worlds. It's as if someone who couldn't see past C wanted to be slightly Erlangish.
No thank you, I will go back to Common Lisp and Haskell any day of the week from this abomination of a language. I've been exploring Clojure (on JVM, CLR and JS) as a practical alternative to CL lately.
I work in C++ and still use inheritance (although I generally prefer composition). One advantage over composition is that it's less typing. For example, if I'm using public inheritance to express a is-a relationship between a base and derived class, all of the public methods of the base are available without having to implement a method in the derived class that just forwards the call to the base class.
I think it's a nice idea. In general, Go seems to provide the language mechanism without the "moral" aspect. In other words, it saves you typing without forcing you to accept the OO paradigm (Liskov substituion etc.).
I would rather see a discussion about performance, platform support, maintainability, governance and do on.
---
[1] someone please create a new programming language where the empty file means "print hello world". Since you can't do any better than that it would once and for all put an end to this stupid benchmark.
While I'm not sure this content should be on the front page, it never was represented as a very deep comparison. I do admit that when I read it through I expected much more.
Also, how do you determine what libraries to use and what projects to invest in these days? Do you always do a full audit of every project? I know I don't have the time, so I use stars on a Github project just like one might use word-of-mouth. If a ton of people think a thing is useful, it's probably at least a little useful. Of course, that kind of thinking can be dangerous (see: javascript ecosystem), but a lot of the time, it's "good enough".
BTW, the paper does cover a tiny bit of information regarding differences in platform support when it covers how Swift deals with concurrency.
Up to a certain point, stars can be useful as it helps you discover projects other people have found of use. But saying project X is better than project Y because it has more stars? What does it show other than more people using X are on github or people using X a more prone to push the star button?
For reference, bootstrap has 100K stars, Linux has 40K...
Because it's a good proxy for how good the language is for writing short, (often) throwaway programs.
Swift is a great joy for iOS development, where compatibility with Objective C makes things smooth, and operation queues are good enough for concurrency. Xcode is a big productivity booster. Good to break Swift's own backward compatibility with new releases to keep innovation, while providing a quick fix tool. Hope it will become a great server language too.
I am not sure this is accurate, because libdispatch has been ported to Linux:
https://github.com/apple/swift-corelibs-libdispatch
They are "...early in the development of this project..." but server-side frameworks use it.
https://github.com/jakerockland/go-vs-swift/issues/8
Thanks for the link! :)
* What is the cost profile of Dispatch relative to the Go scheduler? It seems that Go programs suffer a slow down just to enable multi-threading, for example -- which is not unlike other systems (GHC at some time in the past) that have a similar model.
* Is Dispatch suitable for "millions of threads" ala Erlang? Hundreds of thousands of threads? Dispatch allows you to decide which queue to put things on; but also kind of requires it of you.
I want to know how the languages compare for writing code. Dos one make it easier to write more succinct and correct code? Swift is missing concurrency support until at least the next version, for example.
Swift is built atop LLVM, and it emits code more slowly, but supports many optimizations like hoisting, vectorization, etc. that Go does not currently perform.
Swift feels like a high level language with static typing.
Interestingly, Swift doesn't feel much like Obj-C.
With Go, I'm never sure what I have (pointer? reference? something other and weird?), and can't cast things that should be castable.
Either way, I'd still rather take Haskell or Common Lisp, but I'd gladly take C over Go after having used Go for a highly multithreaded communications server.
> Interestingly, Swift doesn't feel much like Obj-C.
Some things that can be improved:
* It needs more support / packages for Linux (and Windows maybe?). I was using Manjaro Linux, and around November 2016 (dont remember exactly), the existing packages in the AUR didnt work anymore.
* No built-in weak collections.
* No source subfolders for the same project when using the buit-in package manager (I dont know about Go in this matter).
Once Jetbrains Gogland is up and running I will be playing more with Go.
> Note that Go does support a ‘println’ function that does not require import- ing from the standard library; however, this function reports to ‘stderr’, rather than ‘stdout’, and is not guaranteed to stay in the language 19.
Some quick googling found me: http://fernandocejas.com/2016/02/20/how-to-use-optional-on-a... ...
One thing I've found is that once you start using Optionals, they tend to infect (I don't mean this in a bad way) the rest of the code. You start to want Optionals everywhere, and start writing maps/orElses everywhere, which I personally love. The error-handling paradigm changes a little bit (to errors-as-values, which I personally appreciate as well), but some are turned off by having optionals everywhere.
An Optional in Java is just another value, and nothing in practice stops that from being null. I.e. a method that returns an Optional can still return an actual null if poorly implemented. So the major benefit Optionals give you in Java is a typed declaration that the method may return an "empty" value, which is certainly helpful for knowing what cases your client code needs to handle, but it does not guarantee that the method will never return null. It reduces the probability of unhandled null pointers, but it does not eliminate them.
On the other hand, Swift & co. guarantee at compile time that a function which says it returns an Optional will always return an an Optional, never a null, and, better still, if a function says it returns some non-optional value, it still is guaranteed to never return null.
It's great that Java is pushing for this style of code, but I'm afraid without compile-time guarantees of return values it will always feel less powerful there.
You may also enjoy a language that supports the generalization of Optionals, which are Algebraic Data Types (ADTs). Optionals are a single, limited application of ADTs.
Optionals allow you to shift null pointers (and null pointer exceptions) to the type level. So instead of having to worry about a function returning null, you just deal with Optionals whenever a function returns one.
There's another ADT, usually called "Either" or "Result", that allows you to shift exceptions to the type level. You can see from the type of the function what kind of exception it might return, and you deal with exceptions like any other value instead of via a special exception handling mechanism.
For example, in Haskell, by default you can't compare an Optional with the value it wraps: `5 == Just 5` fails. But in Swift this works like you would want.
All that is to say that Options in Swift are a bit nicer than what you could get with pure ADTs. It's a similar story for Swift's syntax-level error handling vs type-level monadic error handling. (The downside of course is that the compiler has to be specially taught about these - but maybe you want that anyways, e.g. for Rust's null-pointer optimization.)
Not having used them much in the past, I rather don't like them, but I'm aware of the fact they could simply take getting used to.
Anyone chime in on whether or not Optionals are really the 'future of good syntax'? Or they are quirk? Or good in some cases, not for others?
Optionals have lots of syntactic sugar because Swift has to deal with reams of native and objective-c code which is ubiquitously nullable, dealing with that without the syntactic sugar would be nothing short of horrendous. This is also an issue with ported/converted API (which is most of them at this point) as the underlying system tends to leverage nullability extensively (it's a very common Obj-C pattern as nil is a message sink), which is usually left shining through by simple conversion rather than wrap every historical API in a converted Swift version.
I love the compile-time optional support now and severely miss it when using languages that don't have the feature. They are brilliant when you are working with all-Swift code or Objective-C code that has been decorated with nullability keywords.
However optionals become a burden when you are working with old Objective-C code that has not been annotated.
Edit: one thing I forgot is the very slow feedback at least when editing Swift. For syntax error notifications to get updated sometimes takes tens of second which can be very confusing.
Great thing about Go is you don't need much of an IDE because it's best to just keep Go in its sweet spot, which is services. LiteIDE works great for Go- small footprint,debugging, enough project management to get by. Just like with everything else about Go, you can get a newbie dev going with the Go toolchain actually producing something that works in hardly any time.
Every single IDE I've used, I've always used another text editor for doing actual programming. It's nice to have a half-decent text editor in the IDE for debugging (see value of variables in the code, looking up and quickly fixing compile errors, etc).. but I don't expect it to be as powerful as vim, emacs, sublime, et.al.
I hate to say it, but It's not production ready yet, at least not for a big project.
There's some tools you can use to diagnose build times, often slow builds are just a line or two that seem to mess with the compiler and can easily be broken apart or written in a different way.
Also, have you tried the Android emulator/Android studio? Last time I did it made me appreciate Xcode and the Apple dev ecosystem a whole lot more.
In that case they're not truly open: Go does not allow you to declare methods on receivers from other packages, which means you can't extend anything which wasn't written by you. Which makes open classes almost useless.
This is how a diamond looks like:
type Base struct {
Foo int32
}
type Child1 struct {
Base
}
type Child2 struct {
Base
}
type Diamond struct {
Child1
Child2
}
func (c *Child1) DoSomething() {
fmt.Println(c.Foo)
}
func (c *Child2) DoSomething() {
fmt.Println(c.Foo)
}
func main() {
c := Diamond { }
c.Foo = 42 // Doesn't Compile
c.Child1.Foo = 10
c.Child2.Foo = 20
c.DoSomething() // Doesn't Compile
}- A stable and pragmatic language that interfaces well with the C world.
- Nice separation of class interface/implementation; ARC is as easy as COM was in Delphi; solid reflection/metaprogramming capabilities (e.g. enumerating properties).
- Compilation is fast enough on my 5y/o laptop.
- A standard library that is so good that people never re-invent it.
- The debugger works really well and is always-on (IntelliJ still has separate "Run" and "Debug" buttons, facepalm).
- IBDesignable/IBInspectable makes writing your own components as nice as it was in Delphi: http://nshipster.com/ibinspectable-ibdesignable/
The points above will probably fall apart with Swift, but right now it's not actually too bad.
True, but perhaps not practically relevant. The pattern matching in Swift isn't much better than what you would get with a tagged C union, which is why no one really uses it very heavily. The Optional type in Swift gets a lot of special compiler support, which is indicative of the fact that the broader language isn't very friendly towards using ADTs to structure data.
But you're right, I should have clarified in my comment that Swift does have a basic degree of support for ADTs.
In what way? Pattern matching in Swift is quite powerful; definitely better than C's switch statement. This blog post is a good overview of its capabilities. https://appventure.me/2015/08/20/swift-pattern-matching-in-d...
> The Optional type in Swift gets a lot of special compiler support
Only the ?. optional chaining operator (and technically the ! postfix operator too, though that could have been implemented in the standard library instead of being special syntax and in fact I'm not really sure why it isn't just a stdlib operator). And probably some logic in the switch exhaustiveness checker so it knows nil is the same as Optional.none. Everything else (including the actual pattern matching against the "nil" keyword) is done in a way that can be used by any type, not just Optionals (nil for pattern matching specifically evaluates to a stdlib type called _OptionalNilComparisonType which implements the ~= operator so it can be compared against any Optional in a switch, and you could do the same for other ADTs too, assuming they have a reasonable definition for nil).
enum in Swift has pattern matching support, can have properties, adopt protocols, etc.
I would disagree with your assessment that "nobody uses them".
Pretty much every large swift project I've seen has had the problems described. Lyft, Uber, Linked In, etc.
You can see it happen pretty plainly with a simple codegen script creating a bunch of dummy code up to X lines of code.
This leads to very long compile times for even simple one line changes.
However I tend to think that what's important about Optionals is the spreading of the notion of thinking critically about failure modes around improper input/output. "Defensive" coding is often considered a mid-range skill (at least I think), and IMO it's because a lot of junior programmers just assume if some method get ObjectThing a, it's going to be there, and not null. I think merely seeing Optional<ObjectThing> a encourages movement in the right direction as far as program correctness, a road that will almost surely tour through concepts and languages that take that kind of correctness more (ex. Swift, Haskell).
Also, Optionals are a light introduction to functors which is nice.
For more concrete actionable stuff please see other comment (https://news.ycombinator.com/item?id=13433335), there are tools that can do this at compile time for java
Why are optionals good and checked exceptions bad? (I'm certainly not implying that one replaces the other - i'm just saying that it they both have an origin of forcing programmers to think about errors)
Exceptions are exceptional; the whole point is that you aren't supposed to have to think of them in normal program flow. Go gets this wrong too, forcing three extra lines of if err != null { for pretty much every single function call -- and with the added bonus of destroying stack context with each return.
I love checked expressions, hate Optionals.
I'll be happy when the language pimps stop trying to make Java look and act like a scripting language.
Edit: I apologize if I've offended anyone with my opinions. I hope this helps. http://bit.ly/1S1943H
Optional is not a functor, in fact it violates the functor laws quite blatantly.
Also, when I wrote "functor" I was thinking of https://wiki.haskell.org/Functor, is that what you're thinking of? In that case, which of the rules does it break? if not, what definition of functor were you thinking of?
Optional.of(whatever).map(identityfunction) will definitely give you back an Optional<Whatever>... Am I missing something fundamental? Also Optional.of(whatever).map(f).map(y) is equivalent in return to Optional.of(whatever).map(f . y)... (of course that's not the java syntax for composing functions but I digress)
val str: Function[String, String] = s => if (s.length > 3) s else null
val num: Function[String, Integer] = s => if (s == null) -1 else s.length
scala> Optional.of("Foo").map[String](str).map[Integer](num)
res12: java.util.Optional[Integer] = Optional.empty
scala> Optional.of("Foo").map[Integer](str.andThen(num))
res15: java.util.Optional[Integer] = Optional[-1]
They generally also break encapsulation (though you can kind of work around/sidestep this by making top-level visible errors more opaque)
Do you have a good reason you'd like to share for liking checked expressions? It seems like the reason you like it is because it's enforced by the compiler -- but I don't think this makes them the right choice, as the compiler could easily also enforce exhaustive handling of an Optional (or sum types like other languages).
If we limit ourselves to the case of Java 1.8, it is a fact that optionals are not checked by the compiler, so there is a solid benefit of using checked exceptions, solely that there is a compile time check of whether the exception was accounted for.
Because they're the best solution for the problem. In Java.
The problem with Java's checked exceptions is misuse; turgid overwrought frameworks which obfuscate. Work closer to the metal. Eschew Spring, JPA, Annotations, aspects (AOP). All terrible ideas (making Java more dynamic) implemented terribly. (Quoting myself: Spring is an exception obfuscation framework.)
If you have to catch an exception you can't handle, revisit the design.
Your linked comment references sum types (Rust, Haskell) and multiple return types (Go). I haven't used those, so I can't comment.
The only solution better than checked try/catch/finally might be proper state machines. Experimenting with that is (far down) on my to do list. I'm also curious about Erlang/Elixir (and CSP in general), which uses a completely different strategy for error handling.
*"... Java 1.8, it is a fact that optionals are not checked by the compiler..."
Java's Optionals are a bad idea implemented badly.
The Correct Answer is using the Null Object pattern. Or just use LISP/Scheme. Also good ideas are intrinsic path expressions (vs method chaining) and pattern matching.
---
Top-thread, someone shared the "composition over inheritance" heuristic. This is more correct. Null Object and composition are like peas and carrots.
You mention "breaking encapsulation". So? Choose composition, make most objects dumb, implement the smart/active parts as yet another Interpreter. Problem solved.
Haskell has bind (>>=) and do-syntax, rust has `and_then`. This is pretty standard with any ADT-supporting language.
> if-let
Most languages with ADT support have good pattern matching that subsumes if-let syntax and allows you to do other things as well. Swift's pattern matching, such as it is, is a bit of a mess.
> guard-let unwrapping
Haskell has `guard` and `MonadFail`, which address the use cases for this in a more principled way. `guard` for boolean matching (like equality) and `MonadFail` for structural matching (like pattern matching on a constructor).
Rust has (?) and `try`, which are syntactic sugar around a simple match/return statement that addresses most use cases of guard-let.
Obviously there are going to be small differences between this implementations, but Swift doesn't really excel here.
> `5 == Just 5` fails.
As it probably should. Supporting such implicit casting could lead to some obvious confusion and buggy behavior. Ideally, the fact that "a == b" typechecks should indicate that "a" and "b" are of the same type.
In Haskell, you would just do `Just 5 == x` if you want to check that something is, in fact, `Just 5`. If that really wasted a lot of space somehow, you can define something like `x ==: y = Just x == y`.
edit: FWIW, there's some overlap between '?' and 'guard let', but not that much. Using hypothetical Rust syntax, they'd overlap in this case:
guard let Some(x) = foo else {
return Err(GettingFoo);
}
let x = foo.ok_or(GettingFoo)?;
Rust copied 'if let' and made it far nicer and more capable by allowing arbitrary fallible pattern matches to be the condition. Then Swift 2 copied that improvement back with 'if case'.
`context(e?)` becomes `e >>= \x -> context(x)`
It's totally fair to point out that Haskell is more principled - users can build Haskell's Maybe, but not Swift's Optional.
> Rust has (?) and `try`, which are syntactic sugar around a simple match/return statement that addresses most use cases of guard-let
They aren't really comparable. Rust's `try!` addresses one case: you have a Result and want to propagate any error to the caller. This is closest to Swift's `try`. But Swift's `guard` is much more flexible: it allows destructuring, boolean tests, multiple assignment, etc., and it also allows arbitrary actions on failure: return, exit(), throw, etc., with the compiler checking that all paths result in some sort of exit.
In practice this is used for all sorts of early-outs, not just error handling. It neatly avoids `if` towers-of-indenting-doom. I think the best analog is Haskell's do-notation.
There's the same tradeoff here. Rust's try! is a macro that anyone can build, while Swift's `guard` is necessarily built into the language. But any Swift programmer will tell you how much they appreciate this feature, unprincipled though it may be. Use it for a while and you become a believer!
> As it probably should. Supporting such implicit casting could lead to some obvious confusion and buggy behavior. Ideally, the fact that "a == b" typechecks should indicate that "a" and "b" are of the same type.
The principled stand! Both languages make the right decision for their use case. Swift's Optionals are much more commonly encountered than Haskell's Maybe (because of Cocoa), and so the language's syntax design optimizes for dealing with Optional. They're more central to Swift than its ADTs.
-- Probably nothing like in Swift
instance (Num a , Integral a) => Num (Maybe a) where
fromInteger x = Just (fromIntegral x)
main = print $ 5 == (Just 5) check :: Int -> Bool
check x = x == (Just x)
main = print (check 5)
Main.hs:7:17: error:
• Couldn't match expected type ‘Int’ with actual type ‘Maybe Int’
• In the second argument of ‘(==)’, namely ‘(Just x)’
In the expression: x == (Just x)
In an equation for ‘check’: check x = x == (Just x)Why in the world would you want those two to be equal when they obviously don't represent the same thing?
That doesn't make sense, not even if they have the exact same memory representation, in which case I'm pretty sure it has been a compromise, which would mean you're still dealing with `null` with some lipstick on it, making that type behave inconsistently with other types in the language.
This kind of misguided convenience is exactly why equality tests in most languages are in general a clusterfuck.
This is the difference between normal people and theoretical computer scientists, summarized in one sentence.
Because I care for intended use, not ontology.
I should have been clear about CPSing things first so I didn't need to eta-expand :)
And they are different because the types say so. By allowing them to be equal, you're implicitly converting one into the other. That's weak typing by definition, a hole in the type system that can be abused.
So why have types at all? Dynamic typing is much more convenient and Clojure deals with nulls by conventions just fine.
Again, it's the use that makes it a safe guard, not its ontology.
>And they are different because the types say so. By allowing them to be equal, you're implicitly converting one into the other.
Which is fine sometimes, when you explicitly need to do it a lot.
>That's weak typing by definition, a hole in the type system that can be abused.
Any examples of how A = Just A can be abused in any meaningful way?
>So why have types at all?
Because I don't believe in the Slippery Slope fallacy, and some types are better than no types at all, but exceptions can be OK too.