Surprised to see a compile error depending upon order of definition/usage of a class in a method!

[Anil Philip]

If I remember correctly, in the 'C' programming language, a program would not compile if the definition was after the declaration.
I thought Java got rid of that.
So I was surprised to see a compile error depending upon order of definition of a class, and its usage in a method!

public class Surprise {
    public static void main(String[] args) {
        A a = new A(); // DOES NOT COMPILE: java: cannot find symbol
        class A {}
    }
}

public class Surprise {
    public static void main(String[] args) {
        class A {}
        A a = new A(); // OKAY!
    }
}

[Mike Simmons]

Yeah, I suppose it feels a little inconsistent.  Java has a well-defined mechanism for telling the compiler where to look for top-level and member classes, based on the package and import statements in the file you're looking at.  But that doesn't apply for local and anonymous classes, where they revert to the old-fashioned define-it-before-you-use-it philosophy.  In practice I've never seen this create a problem.

[Carey Brown]

Anil Philip wrote:So I was surprised to see a compile error depending upon order of definition of a class, and its usage in a method!

public class Surprise {
    public static void main(String[] args) {
        A a = new A(); // DOES NOT COMPILE: java: cannot find symbol
        class A {}
    }
}

I've been writing code this way for many-many moons with no compile error. What version are you using?

[Anil Philip]

Carey Brown wrote:I've been writing code this way for many-many moons with no compile error. What version are you using?

Java 17. Can you please copy-paste my code and see if it runs?

[Campbell Ritchie]

Anil Philip wrote:. . . please copy-paste my code and see if it runs?

Have. Doesn't.

The local class A is out of scope where you are trying to use it.

[Carey Brown]

My mistake, I didn't put it inside a method().

[Anil Philip]

Campbell Ritchie wrote:

Anil Philip wrote:. . . please copy-paste my code and see if it runs?

Have. Doesn't.

The local class A is out of scope where you are trying to use it.

Two-pass compilers were supposed to take care of this issue (at least that's what I remember from Compilers Theory class in college).

[Campbell Ritchie]

I don't know whether javac is a two pass compiler. Order of declaration doesn't usually matter, but local variables, including that local class, have a restricted scope.

[Stephan van Hulst]

The thing is, when designing a good compiler, you always have to make trade-offs between performance and ease of use.

A second compiler pass through a method body would definitely allow you to use a local class before it was declared. But is it really worth it?

In a typical Java application, most lines of code that need to be compiled exist inside method bodies, and of those lines of code, the least of them relate to class declarations. Why force the compiler to analyze all these lines of code a second time, only to make a rarely used feature marginally easier to use?

[Mike Simmons]

[Update: I didn't see Stephan's reply until I posted this, but mine is in a similar vein...]

I imagine the Java compiler does more than one pass of a class in order to compile it, since it needs to be able to know what all the members of a class are in order to be able to reference them out of order.  They presumably build that list of members on one pass, and then on a second pass, use that to determine whether a given method access or field access expression is valid, and if so, what to connect it to.  But apparently, they didn't choose to apply this to allow multiple passes within a method.  Perhaps because they didn't think it would be particularly worthwhile to do so.  I'm sure they could have if they really wanted to - but why would it be useful?

I would argue that aside from technical questions of whether and how they could do this, there's a question of whether it's a good idea.  I think the most relevant concern here, to me, is whether it results in more readable code, or not.  Personally I would find such out-of-order declarations harder to read.  Maybe it's just me - readability is often all about what you're used to, after all.  But I appreciate things that limit the amount of other code I need to read in order to understand a given bit of code.  Declaring things before they are used, for local variables and local types, is one of those things.

[Paul Clapham]

I vaguely remember a discussion here several weeks (months?) ago about how a local variable had to be declared before it could be used later in the method... perhaps because it was used in a lambda expression, but I'm not sure of that. Anyway the issue was (I think) that declaring the variable after it was used made it be out of scope.

[Campbell Ritchie]

Mike Simmons wrote:. . . know what all the members of a class are in order to be able to reference them out of order. . . .

That is a feature I have found lacking in C, which makes Java® easier to write. But, as everybody is saying, that doesn't apply to local variables.
Is that local class a member of the surrounding class? I am not quite sure.

[Mike Simmons]

Campbell Ritchie wrote:Is that local class a member of the surrounding class? I am not quite sure.

No.  A "member" of a class must be directly enclosed by that class.  It can't be contained within something else which is a member of the class.  A local class is directly contained by a code block, for a method or constructor or initializer.  It can't be directly contained by a class.

[Campbell Ritchie]

Thank you. Mike