I'm pretty new with Kotlin and I'm trying to figure out Kotlin's scope functions.
My code looks like this:
with(something) {
when {
equals("test") -> var1 = "test123"
startsWith("test2") -> var2 = "test456"
contains("test3") -> myNullableVar?.let { it.var3 = "test789" }
}
}
So before I entered the third check with the .let function my with function does not need to be exhaustive (I'm not returning something, I'm only doing assignments). In my third check I'm using .let as a null-check ... but only for an assignment of it.var3 (if it is not null). I don't need to return anything while I know that Kotlin's .let function returns the result of the body by standard.
Nevertheless now my with/when needs to be exhaustive otherwise it won't compile anymore.
This got me thinking and trying out different things. I found these ways to solve this issue:
I can add an else to my with/when so it becomes exhaustive but actually I don't need an else and I don't want to use it in this case.
I can add another .let, so it looks like this: myNullableVar?.let { it.var3 = "test789" }.let{} .... but this looks kinda hacky to me. Is it supposed to work like this?
Use If(xy==null){...}else{...} stuff but I thought I can solve this with Kotlin differently
Because I'm new with Kotlin I'm not really sure how to handle this case properly. I would probably just go with my second idea because "it works". Or should I don't use .let for null-checks? Add another empty .let{}? Or did I not get the null-safety concept at all? I feel a little bit lost here. Thanks for any help.
This seems to be an unfortunate combination of features…
A when can be non-exhaustive only when it doesn't return a value. The problem is that the with() function does return a value. And since the when is at the bottom, its value is what gets returned, so in this case it must be exhaustive.
So why doesn't it insist on an else branch even if you omit the "test3" branch? That's because assignments don't yield a value. (They evaluate to Unit, which is Kotlin's special type for functions that don't return a useful value.) If every branch gives Unit, then Kotlin seems* to be happy to infer a default branch also giving Unit.
But the "test3" branch returns something else — the type of myNullableVar. So what type does the when infer? The nearest common supertype of that type and Unit, which is the top type Any?. And now it needs an explicit else branch!
So what to do?
You've found a few options, none of which is ideal. So here are a few more, ditto!
You could return an explicit Unit from that branch:
contains("test3") -> { myNullableVar?.let { it.var3 = "test789" }; Unit }
You could return an explicit Unit from the with():
contains("test3") -> myNullableVar?.let { it.var3 = "test789" }
}
Unit
}
You could give an explicit type for the with(). (It has two type parameters, so you'd need to give both, starting with the type of its parameter):
with<String, Unit>("abc") {
I haven't found a single obvious best answer, I'm afraid…
And to answer your last question: yes, ?.let{ is perfectly idiomatic and common for null checks. In this particular case, replacing it with an if happens to solve the type problem:
contains("test3") -> { if (myNullableVar != null) myNullableVar.var3 = "test789" }
But as well as being long-winded, if myNullableVar is a property and not a local variable, then it opens up a race condition (what if another thread sets it to null in between the test and the assignment?) so the compiler would complain — which is exactly why people use let instead!
(* I can't find a reference for this behaviour. Is there an official word on it?)
Have such code in project:
Cookie CookieCreate(CookiesTypes type)
{
Cookie user_cookie = null;
switch (type)
{
case CookiesTypes.SessionId:
user_cookie = new Cookie("session_id", Convert.ToBase64String(Guid.NewGuid().ToByteArray()));
break;
case CookiesTypes.ClientIp:
HttpListenerContext context = listener.GetContext();
user_cookie = new Cookie("client_ip", context.Request.RemoteEndPoint.ToString());
break;
}
return user_cookie;
}
I understand, that temp variable user_cookie is bad syntax... I've tried to use return in switch-block both in two cases, but I've got an compiler erros, when I tried so:
Pseudo-Code:
case ... :
//some action
return var;
Having a temporary that is set in a case of a switch statement to be returned at the end is not a bad syntax, it is also the only choice if you need to do something on user_cookie for all cases before returning it.
The only problem is see in your code is the lack of a default case which is indeed useful because:
either you can require a default case (so that you do something in that situation)
either the switch should never reach a default case (so you should manage that situation in a special way, for example by throwing an exception)
If you blindly remove temporary variable and return the value directly like you are trying to do, then it gives you a compiler error probably because not all your branches do return something (because you are lacking a default clause or lacking a return after the switch).
Despite the fact that there's nothing inherently wrong with temporary variables, if you really want to avoid it you just need to ensure that all code paths return something.
That means (for example) changing your current return to:
return null;
and having both cases contain:
return new Cookie (whatever);
instead of the assignment.
I often come to this question when coding.
Which of the following examples is a better practice? I am aware that other factors will influence whether one or the other one is better. But in general, what are the advantages of one over the other.
if(object.getA().Value != null) {
return object.getA().Value;
}
return null;
Vs.
string x = string.null;
x = object.getA().Value;
return (x != null) ? x : null;
Here is another similar example:
var a = object.method(x).Value;
var b = object.method(x).Key;
Vs.
var y = object,method(x);
var a = y.Value;
var b = y.Key;
In other words my question is:
Is it better to call a method twice and have one less variable?
or
Is it better to save it into a variable and call the method twice?
Of course if the method results in a lot of processing it might be smart to call it once, but for general cases where the method is not too demanding and the space of the variable is not too big, which one is better and why? or which are the advantages of one or the other?
The difference between them might not make a big difference but I am trying to find better practices and will like to hear the input of some experienced programmers.
Many thanks
Caching the value in a variable is a basic optimization (related to memoizing).
When it becomes truly necessary is if the second call to the function is on the stack a significant percent of the time.
For example, if that second call is on the stack 10% or 20% of the time, then that's how much overall time you can save by caching the first result.
You can keep doing things like that, until the code is as fast as possible.
If I can give an example, ages ago I worked on an app that had code like this:
if (!Done()){
do some stuff
}
....
if (!Done()){
do some other stuff
}
Since Done() was such a short, clean, and simple function to call, it got called a lot.
Never mind that it did a lot including querying a ton of stuff from a DB and throwing most of it away.
Stackshots found the problem instantly.
It depends if you want to be thread safe and if the function could change between calls.
e.g. with
if(object.getA().Value != null) {
return object.getA().Value;
}
return null;
if the implementation of the property getter Value returned a null on the second call you would have a different answer. It could return null on the second call either by implementation of the method or if another thread casused an update between the if and the return statement that made the result of the property null.
This test is actually redundant because you are returning null if it is null. I'm guessing that you meant if (object.getA() != null). Then the previous paragraph still applies but to getA() instead of Value but the if body would throw an null reference exception if getA() returned null on the second call.
So its all down to whether you are worried about the values changing between calls.
General rule: Avoid extra variables (needlessly introduces states).
(Break the rule if calling the function twice adds too much overhead)
This question is a result of the answers to this question that I just asked.
It was claimed that this code is "ugly" because it initializes a variable to a value that will never be read:
String tempName = null;
try{
tempName = buildFileName();
}
catch(Exception e){
...
System.exit(1);
}
FILE_NAME = tempName;
Is this indeed bad practice? Should one avoid initializing variables to dummy values that will never actually be used?
(EDIT -
And what about initializing a String variable to "" before a loop that will concatenate values to the String...? Or is this in a separate category?
e.g.
String whatever = "";
for(String str : someCollection){
whatever += str;
}
)
I think there's no point in initializing variables to values that won't be used unless required by the language.
For example, in C#, the default value for a declared string variable is null, so you're not even gaining anything by explicitly writing it out. It's mostly a style choice, but since strings are immutable, initializing it to something else would actually allocate an extra string in memory that you'd just throw away anyway. Other languages may impose other considerations.
Regarding the string loop, if you change it to a StringBuilder instead you don't even have to think about it.
Edit: removed bits better answered by others.
As a practice, I tend to avoid setting variables to arbitrary values, and instead initialize them to a default value.
i.e.
int x = 0;
string name = "";
bool done = false;
Person randomGuy = null; //or = new Person();
I like this method best because it brings a sense of uniformity to your code while not forcing the next guy that comes along to deal with stuff like: string name = "luke skywalker";
This is all more of a personal preference, so it will vary between programmers.
At the very lest, you should be following the standards that your project as set. You'll have an idea of how the legacy code handles these things, and it's probably best to conform to that so the overall coding style is the same throughout the system.
It depends on the compiler. C# compiler requires that the variable be initialized before using. But the CLR does not have this requirement. At run time the CLR verifies if the variable is initialized or no. If not it will throw a nullreference exception.
In my opinion, it might be more accurate to refer to it as a "code smell" - in the Martin Fowler sense of the word.
I don't think you can change your default initialisation in isolation - it would need to be made in conjunction with other refactoring methods. It is also assuming that you have refactored your code so that you don't need temp variables:
try{
FILE_NAME = buildFileName();
//Do other stuff with file name
}
catch(Exception e){
...
System.exit(1);
}
It also then makes the assumtion that this code segment is only code in the method in which it is contained - i.e. the method only does one thing
When I'm coding I would be concerned that I am using dummy values with temp variables but I would only change when I am finished coding that section and it solves the problem as intended - and only in conjunction with other refactoring steps.
A colleague of mine refactored this code:
private void btnGeneral_Click(object sender, RoutedEventArgs e)
{
Button button = (Button)e.OriginalSource;
Type type = this.GetType();
Assembly assembly = type.Assembly;
string userControlFullName = String.Format("{0}.{1}", type.Namespace, button.Name);
UserControl userControl = (UserControl)assembly.CreateInstance(userControlFullName);
}
to this code:
private void btnGeneral_Click(object sender, RoutedEventArgs e)
{
Button button = (Button)e.OriginalSource;
Type type = this.GetType();
Assembly assembly = type.Assembly;
UserControl userControl = (UserControl)assembly.CreateInstance(String.Format("{0}.{1}", type.Namespace, button.Name));
}
saying that you don't need to create a variable if it is only going to be used once.
My response was that making once-used variables is good practice since it:
functions as and reduces comments (it is clear what "userControlFullName" is)
makes code easier to read, i.e. more of your code "reads like English"
avoids super-long statements by replacing parts of them with clear variable names
easier to debug since you can mouse over the variable name, and in the cases of e.g. PHP programming without debuggers, it is easier to echo out these variable names to get their values
The arguments against this way "more lines of code", "unnecessary variables" are arguments to make life easier for the compiler but with no significant speed or resource savings.
Can anyone think of any situations in which one should not create once-used variable names?
I'm with your opinion in this case. Readability is key. I'm sure that the compiler produces the same executable in both cases, with the compilers as intelligent as they are today.
But I wouldn't claim "always use once-used variables" either. Example:
String name = "John";
person.setName(name);
is unnecessary, because
person.setName("John");
reads equally well - if not even better. But, of course, not all cases are as clear cut. "Readability" is a subjective term, after all.
All your reasons seem valid to me.
There are occasions where you effectively have to avoid using intermediate variables, where you need a single expression (e.g. for member variable initialization in Java/C#) but introducing an extra variable for clarity is absolutely fine where it's applicable. Obviously don't do it for every argument to every method, but in moderation it can help a lot.
The debugging argument is particularly strong - it's also really nice to be able to step over the lines which "prepare" the arguments to a method, and step straight into the method itself, having seen the arguments easily in the debugger.
Your colleague doesn't seem to be consistent.
The consistent solution looks like this:
private void btnGeneral_Click(object sender, RoutedEventArgs e)
{
UserControl userControl = ((UserControl)type.Assembly).CreateInstance(String.Format("{0}.{1}", this.GetType().Namespace, ((Button)e.OriginalSource).Name));
}
I'm completely with you on this one.
I especially use this if a method takes a lot of booleans, ie
public void OpenDocument(string filename, bool asReadonly, bool copyLocal, bool somethingElse)
To me this is a lot more readable:
bool asReadonly = true;
bool copyLocal = false;
bool somethingElse = true;
OpenDocument("somefile.txt", asReadonly, copyLocal, somethingElse);
..than:
OpenDocument("somefile.txt", true, false, true);
Since the programming languages I use generally do not tell me what was null in an exception stacktrace I generally try to use variables so that no more than one item per line can be null. I actually find this to be the most significant limiter of how many statements I want to put on a single line.
If you get a nullpointerexception in this statement from your production logs you're really in trouble:
getCustomer().getOrders().iterator().next().getItems().iterator().next().getProduct().getName()
Although I agree with your thoughts, adding an extra variable can introduce an extra concept in the method and this concept may not always be relevant to the overall goal of the method. So excessive adding of variables can also increase method complexity and reduce legibility. Note the usage of excessive here.
I guess in some cases where it could have an effect on performance. In particular in this example:
for (int i1 = 0; i1 < BIG_NR; i1++)
{
for (int i2 = 0; i2 < BIG_NR; i2++)
{
for (int i3 = 0; i3 < BIG_NR; i3++)
{
for (int i4 = 0; i4 < BIG_NR; i4++)
{
int amount = a + b;
someVar[i1][i2][i3][i4] = amount;
}
}
}
}
... the extra assignment might have a too big impact on performance.
But in general, your arguments are 100% correct.
The both codes are exactly the same. Of course, yours is more readable, maintenable and debuggable, but, if that was the point of your colleague, his code is NOT memory less consumer.
I think it's a judgement call based on how tidy you want you code to be. I also think that both you and your colleague are correct.
In this instance I would side with you colleague based on the code you presented (for performance reasons), however as I said before it does depend on the context in which it will be used and I think your position is perfectly acceptable.
I would point out that creating variables for once used parameters can be pointless, unless they are const variables or things that you need to use in many places.
I would argue that declaring a once used variable could possible create more confusion when you are debugging if there are lots and lots of these, but one here and there is probably fine.
Creating a new variable means one more concept for the reader to keep track of. (Consider the extreme case: int b=a;c=b;) If a method is so complex - in such need of breaking up - that the extra concept is a price worth paying, then you ought to go the whole hog and split it into two methods. This way you get both the meaningful name and the smaller size. (Smaller for your original method: if it's like you say, then people won't typically need to read the auxiliary method.)
That's a generalisation, particularly in a language with a lot of boilerplate for adding new methods, but you're not going to disagree with the generalisation often enough to make it worth leaving out of your style guide.
I'm completely with your colleague in principle, but not in this case.
The problem with throwaway variables is that they introduce state, and more state means the code is harder to understand since you don't know what effects it could have on the program's flow. Functional programming has no variables at all, exactly for this reason. So the fewer variables there are, the better. Eliminating variables is good.
However, in this particular case, where the method ends right after the variable's only use, the disadvantages of having it are minimal, and the advantages you mention are probably more substantial.
Trying hard to come up with an argument against introducing new variables I'd say that when you read the code (for the first time, at least), you don't know if the variable is being used more than once. So immediately you will let your eyes scan down through the code to see if it is used in more places. The longer the function the more will you have to look to see if you can find it.
Thats's the best argument against that I can come up with! :-)
That's how I used to code. Nowadays I tried to minimize intermediate variables. The use of intermediate variables is perfectly fine if it's immutable.
I'm in agreement with the majority here, code readability is key.
It's a rare line count crusader that actually writes highly readable, highly maintainable code.
Additionally, it all gets compiled to MSIL anyway and the compiler will optimise a lot for you.
In the following example, the compiler will optimise the code anyway:
List<string> someStrings = new List<string>();
for (int i = 0; i < 1000; i++)
{
string localString = string.Format("prefix{0}", i);
someStrings.Add(localString);
}
Rather than:
List<string> someStrings = new List<string>();
string localString = string.Empty;
for (int i = 0; i < 1000; i++)
{
localString = string.Format("prefix{0}", i);
someStrings.Add(localString);
}
So there's really no performance reason not to go with it in many cases.
Agreed
"makes code easier to read, i.e. more of your code "reads like English"
I think this is the most important factor as there is no difference in performance or functionality on most moder managed languages
After all we all know code is harder to read than it is to write.
Karl