Point 5 seems like dangerous advice, shortening types (especially floating-point types) without analysis isn't a good idea and especially not a deadly sin.
Changing doubles to floats can give a significant performance boost (mainly on ARM with Neon) but it brings significant limitations on range and accuracy that can lead to subtle bugs if you don't do the analysis (especially if you mostly test it with double precision.)
Very much this. Using floating-point types that have different sizes on different platforms without detailed analysis of the algorithm is a recipe for hard-to-diagnose bugs. Use float, or use double, whichever is most appropriate.
There are some subtle historical reasons why CGFloat is defined the way it is; a reasonable argument could be made that it "should have" been double on all platforms with 20/20 hindsight, but it can't be changed now without breaking binary compatibility.
Regarding the comments on TDD - the important part of TDD is not the unit tests. The important word in 'Test Driven Development' is the 'Driven'. You can actually do TDD without any kind of automated testing, by defining user tests up front in English, and then manually testing these when hitting milestones. The automated unit tests just make this process more streamlined, but are very much not the key aspect of TDD.
... and are prone to human error. While you could do manual tests before every check-in, this is prone to error to the point that I would not trust it. For any a project of any significant size such a thing would inevitably lead either to steps being missed or the entire thing being skipped altogether.
100%. Automated tests are far superior to manual. The point, however, is that TDD is not the same thing as unit tests. Too many people add some unit tests to their code and then claim they are doing TDD. Too many people try to do TDD by having 100% unit test coverage. It's very important to get the two concepts straight.
Automated tests are only superior when it is too intensive to do manually e.g. integration tests or serialization tests.
But otherwise especially for iPhone apps manual testing is far, far superior. It will always pick up issues with the UI that automated testing never will. And the UI issues are the most important ones.
Oh man, am I guilty of not using dot syntax consistently. I actually try not to use it because it is so out of place with the rest of the code, but it's a leftover habit from other languages, and old habits die hard.
Objective-C dot syntax is fucktarded. The dot is a structure-member indicator; why are you using it on a value of pointer type?
The thing is, Smalltalk has a perfectly good getter/setter convention that's almost as compact and straightforward: for a property called 'foo' the getter is called 'foo' and the setter is called 'foo:' (takes one arg). Apple could have used this convention in its method synthesizer (they opted to make the setter the somewhat more awkward 'setFoo:') and everybody would be clear on what was a struct member access and what was a message sent to an object. But no.
Cocoa is pretty good, but like all frameworks, it definitely has its downsides. Point #1, unfortunately, seems to be something that the author has a hangup about.
I find that using a single XIB for a view controller at a time works the best for me, and maintains some semblance of MVC. I work on a relatively large app, with more screens than usual, and I never find it hard to figure out view controller transitions or anything like that.
I think it is worth mentioning that there is Cocoa and there is Cocoa Touch: supposedly saner and cleaner version. I am only familiar with the Cocoa Touch so I cannot compare myself.
I hate Storyboards - I can't even come close to manipulating views/subviews the way I want to with them.
I love dot syntax. Writing myArray.count is a lot easier than [myArray count] and then when messages start getting nested it bothers me visually - ie [[myView subviews] count]. Of course, I love messages and the syntax used for calling functions too.
Objective-C is my first real language though, so what do I know.
I have to say I very much prefer [myArray count] to myArray.count since I think of it as a method instead of a property though I agree using dot syntax for synthesized class properties is preferable.
Point 6 is a rather uninformed opinion. The reason the Keychain and Sandbox API are written in C is because they are designed to be used straight up by C programs without calling in obscure Obj-C land. Have you ever wondered why Mercurial and Subversion are able to put your credentials in Keychain? This is why. Lower-level, general purpose Mac OS APIs have always been written in C, and they should.
I switched to SSKeychain (https://github.com/soffes/sskeychain) after using SFHFKeychainUtils for a while because I think it's cleaner and more regularly updated.
I always ask myself why I'm thinking about these things in the first place. Its a failure of the language, in my opinion, a deep design flaw that isn't being recognized.
Why do things need to be synthesized in this day and age. I know why technologically, but it doesn't have any actual use in real life. Its just window dressing and something special the developer needs to think about, with no real solid gain other than to entrench a platform-specific methodology into the language, and thus ensure developer mindshare is captured and maintained ..
Is there any compelling reason that we know of for Apple using Objective-C in their development tools? In all seriousness, as a programming language, what does it bring to the table that C, C++, or some scripting language do not?
(I ask this in good faith--not trying to troll. I am genuinely curious.)
Amazing that this isn't voted higher. Regardless of Objective-C's virtues, this the real reason. The decision was made back in the late '80s for NeXTStep, and it just hasn't been changed since then. Apple was kinda moving toward Java in the early days of Mac OS X — WebObjects was ported to it, and it was a peer language for Cocoa up to Mac OS X 10.4 or something — but Mac users hated Java, as it was dog slow at the time, so Objective-C won by inertia.
Objective-C has the dynamic properties of a language like Smalltalk, but you can also drop down to C and write fast imperative code.
Besides the raw capabilities of the language, the community of Objective-C developers fit very well with the language. Where Ruby would use an abbreviation, ObjC prefers long explicit names. And the interleaved method name/arguments makes the code readable in a very specific and refined aesthetic. It isn't obvious to everyone, and it isn't a natural way to write some kinds of programs. But for writing fast and highly-interactive UIs, it remains a high water mark.
The best advice is: try going through one of the books by Aaron Hillegass to get a sense of the serenity of writing proper Objective-C/Cocoa code.
Look up Smalltalk for some hints. I'm not an expert on Obj-C or Smalltalk but Obj-C borrowed pretty heavily in some respects, as I understand it. It is object-oriented, but it is a substantially different paradigm from the C++/Java family.
As another hint, Ruby was pretty heavily influenced by Smalltalk as well. Think about stuff like responds_to? and whatnot, where you're sending messages to objects and objects can actually intercept and react to those messages.
The comparison to C++ is pretty illustrative, if only as a point of contrast; C++ is certainly not the only possible approach to OO in a C-like language. Alan Kay, the guy who did Smalltalk and OO, doesn't particularly approve of the C++ approach to OO.
In practical terms, though, I think it's somewhat historical, to the extent that any programming language's usage is or is not historical. :)
If you're going to expose an OO API then you'll need an object system of some kind (see also: COM and GObject). Objective-C is really just an object system, along with a minimal amount of syntactic sugar added to C to simplify its use.
Obviously people can argue for ever over the merits of one design decision over another but as someone who has programmed in Java for more than a decade, and has used Scala and Clojure commercially, I'd say that Objective-C does have some attractive qualities:
1. The object model is similar in complexity to Java (i.e. no multiple inheritance, templates, etc. that fill C++ with corner cases).
2. It is C but with an object oriented message passing layer on-top. You can think of an Objective-C program as a bunch of C programs that have a more loose coupled way of talking to each other. This gives it a balance between the abstraction of a high-level language and the performance and control of C that is different from Java and most scripting languages where you need to use a native-API to talk to C. Whether this is objectively good or not is debatable but it certainly supports Apple's strategy. Personally I find it liberating.
3. The dynamism and looseness of the language make code transformation tools much harder to write - which is why XCode is only now starting to approach Eclipse on refactorings and completion. The upside is the dynamism enables a bunch of things whose equivalents require bytecode manipulation in java - e.g. property observers, Core Data synthesized accessors, Undo proxies etc.
Java is evolving slowly to make these kinds of thing easier, and Objective-C is evolving slowly to a tighter language that's more amenable to automated transformation.
These are just a few reasons. I think it's also fair to say that Objective-C was pretty crude and was 'behind' even as recently as 2007, but the rate of improvement since then has been high and has brought it up to date. (Declared properties, blocks, ARC, GCD, are pretty major steps into the modern era)
There are definitely a lot of rough edges that still remain and there's a lot more to learn than say a scripting language, plus a very different philosophy to get to grips with by comparison with the java family but once you do know it, it's a powerful language with some great strengths.
I'm very interested to see what happens to it over the next few years.
C pretty much still Rules the System Layer. Its never going to go away for as long as its easy, and efficient, to write very good C code which performs well. In spite of the hate, there is a lot of really good C code out there still running, still working, still burning up the market. I'm pretty sure there's nary a system image which doesn't, eventually, get itself operating per the rules of C, for the most part, somewhere ..
That said, for all your very valid points about Objective-C, the same (essentially) can be said of Lua, and the Lua VM, for example.
As a mobile developer, I'm no longer interested in Objective-C - its only an Apple Language. But I can take the Lua VM and put it on all the other machines, host-wise/parasitically, and create my own internally ordered Framework which runs on all Platforms, and still gain a lot of the benefits of a re-evaluation of 'language simplicity' versus programmer effectiveness.
After 4 years of Mobile development on iOS and Android, where multiple projects have blossomed organically into unwieldy godawful trees of complexity which prove, every day, even more difficult to turn over to other programmers, themselves creating massive WordSpaceCollections: ofCode.to_BeMaintained((Some*)Way) || Other { NSLog("grr..", &etc} ;
On a Drama Scale, it goes like this:
"Oh, Android NDK/SDK, how you have blossomed to being something I regret I am not putting into the trash in the early days. XCode, you %(#@&% Asshole piece of software, Why I Gotta Download CMDLineTools just to get work done"
..
"SublimeText2, factory settings .. Open Folder->".lua files", build and distribute for MacOSX, Linux, Windows, iOS, Android, and still only need to maintain one codebase.
This seems like a long way of saying that you don't like Objective-C because it's not cross platform, and that you've decided instead to write all your mobile code in Lua and to maintain your own abstraction layer onto MacOSX, Linux, Windows, iOS, and Android.
The question I was responding to was simply whether there are compelling reasons for Apple to use Objective-C.
Supporting cross platform native development is clearly not a strategic goal for them.
I think I'd add another reason - which is that they are in control of the evolution of Objective-C.
I'm really more trying to point out that there is a great way to escape the trap being laid for you by Apple and their plans for Objective-C, which is indeed to keep the language in their own privileged domain.
And it really is important enough that anyone considering learning Objective-C today, or even using it, know that there is a way out of it: roll your own walled garden and plant what you like within it, on any platform you can.
I would be willing to wager a small bet that says that the scripted-VM-glommed-in-a-web-of-libs approach to the Platform wars will become more and more a key survival strategy in software development over the next 2 years.
The OS, and indeed Distributions are dead; long live the new King, VM-managed library bundling..
I'm not sure what you mean by 'keep the language in their own privileged domain' actually means, but I do agree that it is Apple's strategy to invest in Objective-C above other languages on its own platform, and that they have no investment in making a cross-platform framework other than HTML5.
I don't see why you describe Apple's approach as a 'trap'. They are providing a lot of software components that save effort for those who use them. The results are platform specific, but everyone who uses them knows that and chooses to make that tradeoff intentionally.
What you describe as "VM-managed library bundling" sounds a lot like "building your own platform out of open source parts and maintaining a compatibility layer to your target platforms".
That strategy works for a few of the largest most-well resourced projects - e.g. the browsers Firefox, Chrome, Safari, Opera etc, plus the Adobe Suite, and even these draw criticism for the results not being as good as they could be if they focussed on one platform.
Something like this works on the web too - where people assemble a 'platform' out of javascript libraries - because the base platform simply doesn't provide enough.
I don't see it being a viable strategy for a small team or an individual developer trying to build native applications though because of the amount of time you'll spend keeping the compatibility layer up-to-date with the rapidly changing underlying platforms.
Run with the pack and eventually become de-marginalized on the slippery platform being controlled by Apple, or work a bit harder and gain traction on every platform you can. This has always been a strategic decision for developers, large and small groups, and both paths have their merits and pitfalls. Neither is a guarantee of success.
Also, its not as hard to "keep the compatability layer" up to date as you state .. you need to do it anyway, if you use the vendor-provided native tools. If you can do that, you can take it one step further, and maintain a very productive outer-shell over the trap-tools. That can be a very good strategy, or a poor one; all the examples you provide may be the larger failure cases, but there are smaller success cases underlooked in your argument .. MOAI, Love2D, GamePlay .. these are all coming along to eat the Native Development Devotee's lunch ..
Why it's used by Apple is inertia, what it brings to the table that C and C++ don't is balance.
Obj-C tends to bring the best of both worlds, while making slight tradeoffs. It brings the benefits of compile time type checking, while retaining dynamic typing at runtime. Try adding methods at runtime to a C++ object.
Obj-C is reasonably good at a number of things and integrates very cleanly with C++ and C.
Unit testing (and by extension, TDD) is easy in Objective C: Just test your models. Your models are (should be) the only thing that makes your app unique.
WRT point 3: Xcode will now allow dot-syntax access to non-property getter methods, however its arguably clearer to use dot syntax only for properties since they're declared and documented separately from old-style getter methods.
There's also a slippery slope argument, since something like myMutableArray.removeAllObjects will compile (albeit with a warning).
45 comments
[ 4.1 ms ] story [ 81.3 ms ] threadChanging doubles to floats can give a significant performance boost (mainly on ARM with Neon) but it brings significant limitations on range and accuracy that can lead to subtle bugs if you don't do the analysis (especially if you mostly test it with double precision.)
There are some subtle historical reasons why CGFloat is defined the way it is; a reasonable argument could be made that it "should have" been double on all platforms with 20/20 hindsight, but it can't be changed now without breaking binary compatibility.
But otherwise especially for iPhone apps manual testing is far, far superior. It will always pick up issues with the UI that automated testing never will. And the UI issues are the most important ones.
The thing is, Smalltalk has a perfectly good getter/setter convention that's almost as compact and straightforward: for a property called 'foo' the getter is called 'foo' and the setter is called 'foo:' (takes one arg). Apple could have used this convention in its method synthesizer (they opted to make the setter the somewhat more awkward 'setFoo:') and everybody would be clear on what was a struct member access and what was a message sent to an object. But no.
I find that using a single XIB for a view controller at a time works the best for me, and maintains some semblance of MVC. I work on a relatively large app, with more screens than usual, and I never find it hard to figure out view controller transitions or anything like that.
I love dot syntax. Writing myArray.count is a lot easier than [myArray count] and then when messages start getting nested it bothers me visually - ie [[myView subviews] count]. Of course, I love messages and the syntax used for calling functions too.
Objective-C is my first real language though, so what do I know.
Why do things need to be synthesized in this day and age. I know why technologically, but it doesn't have any actual use in real life. Its just window dressing and something special the developer needs to think about, with no real solid gain other than to entrench a platform-specific methodology into the language, and thus ensure developer mindshare is captured and maintained ..
(I ask this in good faith--not trying to troll. I am genuinely curious.)
Besides the raw capabilities of the language, the community of Objective-C developers fit very well with the language. Where Ruby would use an abbreviation, ObjC prefers long explicit names. And the interleaved method name/arguments makes the code readable in a very specific and refined aesthetic. It isn't obvious to everyone, and it isn't a natural way to write some kinds of programs. But for writing fast and highly-interactive UIs, it remains a high water mark.
The best advice is: try going through one of the books by Aaron Hillegass to get a sense of the serenity of writing proper Objective-C/Cocoa code.
As another hint, Ruby was pretty heavily influenced by Smalltalk as well. Think about stuff like responds_to? and whatnot, where you're sending messages to objects and objects can actually intercept and react to those messages.
The comparison to C++ is pretty illustrative, if only as a point of contrast; C++ is certainly not the only possible approach to OO in a C-like language. Alan Kay, the guy who did Smalltalk and OO, doesn't particularly approve of the C++ approach to OO.
In practical terms, though, I think it's somewhat historical, to the extent that any programming language's usage is or is not historical. :)
1. The object model is similar in complexity to Java (i.e. no multiple inheritance, templates, etc. that fill C++ with corner cases).
2. It is C but with an object oriented message passing layer on-top. You can think of an Objective-C program as a bunch of C programs that have a more loose coupled way of talking to each other. This gives it a balance between the abstraction of a high-level language and the performance and control of C that is different from Java and most scripting languages where you need to use a native-API to talk to C. Whether this is objectively good or not is debatable but it certainly supports Apple's strategy. Personally I find it liberating.
3. The dynamism and looseness of the language make code transformation tools much harder to write - which is why XCode is only now starting to approach Eclipse on refactorings and completion. The upside is the dynamism enables a bunch of things whose equivalents require bytecode manipulation in java - e.g. property observers, Core Data synthesized accessors, Undo proxies etc.
Java is evolving slowly to make these kinds of thing easier, and Objective-C is evolving slowly to a tighter language that's more amenable to automated transformation.
These are just a few reasons. I think it's also fair to say that Objective-C was pretty crude and was 'behind' even as recently as 2007, but the rate of improvement since then has been high and has brought it up to date. (Declared properties, blocks, ARC, GCD, are pretty major steps into the modern era)
There are definitely a lot of rough edges that still remain and there's a lot more to learn than say a scripting language, plus a very different philosophy to get to grips with by comparison with the java family but once you do know it, it's a powerful language with some great strengths.
I'm very interested to see what happens to it over the next few years.
That said, for all your very valid points about Objective-C, the same (essentially) can be said of Lua, and the Lua VM, for example.
As a mobile developer, I'm no longer interested in Objective-C - its only an Apple Language. But I can take the Lua VM and put it on all the other machines, host-wise/parasitically, and create my own internally ordered Framework which runs on all Platforms, and still gain a lot of the benefits of a re-evaluation of 'language simplicity' versus programmer effectiveness.
After 4 years of Mobile development on iOS and Android, where multiple projects have blossomed organically into unwieldy godawful trees of complexity which prove, every day, even more difficult to turn over to other programmers, themselves creating massive WordSpaceCollections: ofCode.to_BeMaintained((Some*)Way) || Other { NSLog("grr..", &etc} ;
On a Drama Scale, it goes like this:
"Oh, Android NDK/SDK, how you have blossomed to being something I regret I am not putting into the trash in the early days. XCode, you %(#@&% Asshole piece of software, Why I Gotta Download CMDLineTools just to get work done"
..
"SublimeText2, factory settings .. Open Folder->".lua files", build and distribute for MacOSX, Linux, Windows, iOS, Android, and still only need to maintain one codebase.
tl;dr - You can't do that with Objective-C.
The question I was responding to was simply whether there are compelling reasons for Apple to use Objective-C.
Supporting cross platform native development is clearly not a strategic goal for them.
I think I'd add another reason - which is that they are in control of the evolution of Objective-C.
And it really is important enough that anyone considering learning Objective-C today, or even using it, know that there is a way out of it: roll your own walled garden and plant what you like within it, on any platform you can.
I would be willing to wager a small bet that says that the scripted-VM-glommed-in-a-web-of-libs approach to the Platform wars will become more and more a key survival strategy in software development over the next 2 years.
The OS, and indeed Distributions are dead; long live the new King, VM-managed library bundling..
I don't see why you describe Apple's approach as a 'trap'. They are providing a lot of software components that save effort for those who use them. The results are platform specific, but everyone who uses them knows that and chooses to make that tradeoff intentionally.
What you describe as "VM-managed library bundling" sounds a lot like "building your own platform out of open source parts and maintaining a compatibility layer to your target platforms".
That strategy works for a few of the largest most-well resourced projects - e.g. the browsers Firefox, Chrome, Safari, Opera etc, plus the Adobe Suite, and even these draw criticism for the results not being as good as they could be if they focussed on one platform.
Something like this works on the web too - where people assemble a 'platform' out of javascript libraries - because the base platform simply doesn't provide enough.
I don't see it being a viable strategy for a small team or an individual developer trying to build native applications though because of the amount of time you'll spend keeping the compatibility layer up-to-date with the rapidly changing underlying platforms.
Also, its not as hard to "keep the compatability layer" up to date as you state .. you need to do it anyway, if you use the vendor-provided native tools. If you can do that, you can take it one step further, and maintain a very productive outer-shell over the trap-tools. That can be a very good strategy, or a poor one; all the examples you provide may be the larger failure cases, but there are smaller success cases underlooked in your argument .. MOAI, Love2D, GamePlay .. these are all coming along to eat the Native Development Devotee's lunch ..
Obj-C tends to bring the best of both worlds, while making slight tradeoffs. It brings the benefits of compile time type checking, while retaining dynamic typing at runtime. Try adding methods at runtime to a C++ object.
Obj-C is reasonably good at a number of things and integrates very cleanly with C++ and C.
Is he aware that Objective-C can be used on non-Apple platforms?
There's also a slippery slope argument, since something like myMutableArray.removeAllObjects will compile (albeit with a warning).