Prev: Capture image from webcam in java
Next: Oracle insert current date and time
From: Lethal Possum on 16 Feb 2010 14:11
Hi Lew,

On 16 fév, 17:26, Lew <no...(a)lewscanon.com> wrote:
> Lethal Possum wrote:
> >> Let's say I have a class B that extends a class A. Now I have an
> >> instance of B and I need to "clone" it but only as an instance of A,
> >> not B. I can't just cast my instance of B into A, I need a new
> >> instance of A, and only A. I need that the new object's getClass()
> >> method to return A.class.
>
> >> Is there an easy way to do this? I'd prefer not to copy each field one
> >> by one as there is many of them and maintenance would be difficult
> >> (i.e. how do I make sure someone adding a field to class A will update
> >> my method accordingly).
>
> >> I know I could probably achieve this by reflection, iterating on every
> >> field of class A, but some fields of A need to be copied in a specific
> >> way. This is already done properly by the clone() method of A so I
> >> would really like to leverage that code.
>
> >> I am not sure my problem is very clear so I wrote a short piece of
> >> code to demonstrate it. I also included all the solution that I
> >> already know not to work:
>
> >> == Start of code ==
>
> >> public class A implements Cloneable {
>
> >>   private boolean _copy = false;
>
> The Java naming conventions call for no underscore in this variable name.

Noted.

>
> It initializes to 'false', then you set it to 'false'.  This is redundant but
> some consider it useful for internal documentation and don't mind the extra
> assignment of 'false'.
>

Your point is valid. In this case I don't mind as this code has no
purpose except to be an example.

>
> >>   public boolean isCopy() {
> >>     return _copy;
> >>   }
>
> >>   public Object clone() throws CloneNotSupportedException {
> >>     A a = (A)super.clone();
> >>     a._cloned = true;
> >>     return a;
> >>   }
>
> >> }
>
> >> I just noticed a typo in my code, "a._cloned = true;" should be
> >> "a._copy = true;".
>
> >> public class B extends A {
>
> >>   public A test1() throws Exception {
> >>     A a = (A)this;
>
> Upcasts are superfluous.  'this' already /is-an/ 'A', and doesn't lose its
> 'B'-ness by having an 'A' pointer reference it.
>
> >>     return (A)a.clone();
> >>   }
>
> >>   public A test2() throws Exception {
> >>     A a = (A)clone();
>
> Same remark - upcasts are superfluous.
>
> >>     return a;
> >>   }
>
> >>   public A test3() throws Exception {
> >>     A a = (A)super.clone();
>
> Once again the cast is superfluous.  It does nothing to upcast an instance to
> a type that it already is.
>
> This fails for the same reason that the call to 'super.clone()' in 'A'
> succeeds.  From the Javadocs for 'clone()' (which surely you have read while
> researching this question - right?):
>
> >  this method creates a new instance of the class of this object
>
> Since the class of this object is 'B', the cloned instance is a 'B'.
>
> You can't very well rely on this behavior in 'A' and simultaneously wish it
> wouldn't work that way in 'B'.
>
> >>     return a;
> >>   }
>
> >>   public A test4() throws Exception {
> >>     A a = (A)clone();
> >>     return (A)a.clone();
>
> Not only is the upcast superfluous, and not only does 'a' continue to point to
> a 'B' instance (casts don't change the runtime type of an object), but now
> you've created two instances of 'B' and thrown one away.
>

I completely understand all the reasons why upcasting the result of a
B.clone() will never give me an instance of class A. That what I meant
by "all the solutions that I already know not to work". I should
probably have said "understand not to work".

>
> >>   }
>
> >> }
>
> >> public class test {
>
> >>   public static void main (String[] args) throws Exception {
> >>     B b = new B();
> >>     A a1 = b.test1();
> >>     System.out.println(a1.getClass().toString() + " copy=" +
> >> a1.isCopy());
> >>     A a2 = b.test2();
> >>     System.out.println(a2.getClass().toString() + " copy=" +
> >> a2.isCopy());
> >>     A a3 = b.test3();
> >>     System.out.println(a3.getClass().toString() + " copy=" +
> >> a3.isCopy());
> >>     A a4 = b.test4();
> >>     System.out.println(a4.getClass().toString() + " copy=" +
> >> a4.isCopy());
> >>   }
>
> >> }
>
> >> == End of code ==
>
> >> All that I get is:
>
> >> class B copy=true
>
> Not possible.  You have four 'println()' calls.  You must have gotten four
> lines of output.

Sorry if I was not clear, what I meant is that I get 4 times:

class B copy=true

>
> >> When I would like to get:
>
> >> class A copy=true
>
> The presence of the '_copy' variable (whose name violates the naming
> conventions, btw) is against the spirit of 'clone()', but never mind.  As the
> Javadocs say, for the clone to be value-equal "is not an absolute requirement."
>
> 'clone()' is the wrong method for what you are asking.  It is designed to
> return an instance of the cloned object's class.  You are trying to violate
> that contract.  Create your own method in 'A' to make a copy that instantiates
> its own 'A' instance and copies over the desired state without using 'clone()'.

Creating a method that copies over the desired state will be my last
resort because the practical maintenance of the method will be very
difficult: the actual classes have dozen of fields and other
developers work in theses classes too. When one developer had a new
field, I can count on him (most of the time) to update the clone
method for example but I think there is very little chance that he
will know to update my copy method. Hence my hope to find a solution
based on clone(). Or another solution that I didn't think of but would
be self-maintaining.

>
> --
> Lew

Thanks

Tom
From: Lew on 16 Feb 2010 15:01
According to Lethal Possum:
>> Is there an easy way to do this ['clone()' a subtype of 'A'
>> and lose the subtype information]?

Thomas Pornin wrote:
> Generally speaking, no.
>
> In Java, the implementation of a class is the guardian of what can be
> part of an instance of that class. To get an instance of A you have to
> go through a constructor of A. If A does not override clone(), then
> calling super.clone() from an instance of B will fallback on
> Object.clone(), which will build a new instance of B, not a new instance
> of A. Basically, to get what you want, you will need a bit of
> cooperation from class A.

In the OP's case 'A' follows best practice and its override calls
'super.clone()', presumably all the way up the ladder to 'Object'.

> If class A cooperates (i.e. you are developing the source code of A as
> well), then the simplest way seems to add a A.dup() method which returns
> a new instance of A (created with an explicit 'new A()') and manually
> filled with the proper data. Otherwise, you could probably try to abuse
> serialisation (serialise your instance of B into some bytes, and
> cut&glue it into a proper serialisation for A) but this is hardly "easy"
> and I guess that there are a huge number of pesky details to take into
> account.

Could the OP use a private 'dup()' method in 'A' to implement 'A#clone()'
without calling on A's 'super.clone()'?

There are a host of problems with this but it could work for the specialized
situation, perhaps. 'A' would have to be a direct descendant of 'Object' to
have a hope of getting away with it, I think.

Personally, I would just use a conventional 'clone()'. What does it matter to
the caller who's looking for an 'A' that the type is actually 'B extends A'?
That is the usual pattern for hidden implementation classes. The details of
the implementation class are inaccessible to the caller who got an 'A', and
the caller interacts with the type just as if it were any other kind of 'A'.

--
Lew
From: markspace on 16 Feb 2010 16:16
Lethal Possum wrote:
>
> Creating a method that copies over the desired state will be my last
> resort because the practical maintenance of the method will be very
> difficult: the actual classes have dozen of fields and other
> developers work in theses classes too. When one developer had a new
> field, I can count on him (most of the time) to update the clone
> method for example but I think there is very little chance that he
> will know to update my copy method.


Try to make the copy method as obvious as possible:

public class CopyMe {

int field1;
int field2;

public class CopyMe() {}

public class CopyMe( CopyMe copy ) {
this.field1 = copy.field1;
this.field2 = copy.field2;
}
}

class Child extends CopyMe {}


Making your "copy method" a constructor gives a bigger hint to a future
developer that intervention is needed. Really this is not any worse
than overriding equals(). Developers have to know to update that when
they add a field, and they should know to update a constructor to
initialize fields too.

Adding big comments like /* ANY NEW FIELDS MUST BE INITIALIZED HERE */
doesn't hurt either.
From: Patricia Shanahan on 16 Feb 2010 20:30
Lethal Possum wrote:
> Hello everyone,
>
> Let's say I have a class B that extends a class A. Now I have an
> instance of B and I need to "clone" it but only as an instance of A,
> not B. I can't just cast my instance of B into A, I need a new
> instance of A, and only A. I need that the new object's getClass()
> method to return A.class.
....

The best way I know of to do this is to provide a constructor in A that
takes an A reference as argument and initializes the new A to match the
A aspects of the argument A. That has already been suggested.

So far, I have never needed to do this. The problem you are trying to
solve may be solvable some other way, or there may be some change in
your classes that would remove the need.

Can you explain the higher level objective? Why do you want to do this?

Patricia
First  |  Prev  | 
Pages: 1 2
Prev: Capture image from webcam in java
Next: Oracle insert current date and time