Reputation: 863
Speaking to people about this around the office made me realise that without the context of how I got here, their response is always "You're doing it wrong", so here's the context:
I've got to deal with crappy objects created by another team using ICE (https://doc.zeroc.com/display/Ice37/Ice+Overview) and wanted to write a utility for the boiler plate code I've been stuck writing.
One of the objects I've got to deal with is a container for floats and strings. I just want the float values.
public class FloatWrapper {
public String StringValue;
public float FloatValue;
public FloatWrapper(float f) {
this.FloatValue = f;
}
public FloatWrapper(String s) {
this.StringValue = s;
}
public FloatWrapper(float f, String s) {
this.FloatValue = f;
this.StringValue = s;
}
}
This container is delivered to me in a 2D array so I want a float[][]. Let's start with 1 dimension
FloatWrapper[] wrappers = new FloatWrapper[] {
new FloatWrapper(1.0f),
new FloatWrapper(2.0f)
};
Float[] asFloats = Arrays.stream(wrappers)
.map(f -> f.FloatValue)
.toArray(Float[]::new);
That was easy. Let's try a 2D array.
// Make a lovely method for what I have above
public Float[] convert1Darray(FloatWrapper[] wrappers) {
return Arrays.stream(wrappers)
.map(f -> f.FloatValue)
.toArray(Float[]::new);
}
public static Float[][] convert2Darray(FloatWrapper[][] wrappers) {
return Arrays.stream(wrappers)
.map(f -> convert1Darray(f))
.toArray(Float[][]::new);
}
...
// Create a 2D array
FloatWrapper[][] wrappers2d = new FloatWrapper[][] {
{ new FloatWrapper(1.0f), new FloatWrapper(2.0f) },
{ new FloatWrapper(3.0f), new FloatWrapper(4.0f) }
};
Float[][] floats2d = convert2Darray(wrappers2d);
Boom!
It'd be nice to have primitive floats (it just would, ok!) The wizened among you can see where I'm going...
Turns out you can't use primitive floats very easily in the 'toArray' call. Apache commons will sort that
public static float[] convert1DprimitiveArray(FloatWrapper[] wrappers) {
return ArrayUtils.toPrimitive(
Arrays.stream(wrappers)
.map(f -> f.FloatValue)
.toArray(Float[]::new));
}
public static float[][] convert2DprimitiveArray(FloatWrapper[][] wrappers) {
return Arrays.stream(wrappers)
.map(f -> convert1DprimitiveArray(f))
.toArray(float[][]::new);
}
.... wait a minute. Why can I use 'toArray(float[][]::new)' but not 'toArray(float[]::new)' ?
The argument floating around the office right now is that the 'float[]::new' call is illegal because you can't call 'new' on a primitive.
But that's not what it's doing. That's just shorthand for the IntFunction
Let's break it down
This:
Arrays.stream(wrappers1d).map(f -> f.FloatValue).toArray(Float[]::new);
is the equivalent of this:
Arrays.stream(wrappers1d).map(f -> f.FloatValue).toArray(new IntFunction<Float[]>() {
@Override
public Float[] apply(int size) {
return new Float[size];
}
});
So, is it that you can't create something of type IntFuntion ??
Nope. Totally allowed:
IntFunction<float[]> happyToCompile = new IntFunction<float[]>() {
@Override
public float[] apply(int size) {
return new float[size];
}
};
Executing that function leaves you with the equivalent of
new float[]{ 0.0f, 0.0f };
So, my half-educated guess is that there's some type inference going on from the IntFunction which works when you have a 2d array but not a 1d array. But, 'could' it work?
Upvotes: 11
Views: 5497
Reputation: 19910
That error occurs, because there exists no wrapping stream for float
. So you can't work with FloatStream
or something similar.
So when you analyse what Stream.toArray()
really does you get something like this:
Return an array
T[]
which is holding every object of typeT
in this stream
The problem comes with the T
because you have a stream containing floats
, stored in their WrapperClass Float
(because primitives cannot be used with generics).
Now when you try to convert a Float
to a float
, java does it for you with auto-unboxing
. The bad thing is, it does not the same for arrays.
In java an array of a primitives can not directly be converted to it's specific Wrapper type.
So Float[] floats = new float[10];
gives a compile error.
Now you ask yourself: Why does float[][]
work then?. Well, if you look at it then you see that float[]
is an Object
and a stream of float
-arrays is valid, because they can be referenced as an Object. Thats why float[][]::new
works. It matches the description from earlier: Return an array T[] which is holding every object of type T in this stream because float[]
actually is of type T
and an T[]
is then valid too.
Upvotes: 2
Reputation: 93902
Why can I use 'toArray(float[][]::new)' but not 'toArray(float[]::new)' ?
As you said, Arrays.stream(wrappers).map(f -> f.FloatValue)
is a Stream<Float>
so toArray(float[]::new)
does not work as you are required to provide an IntFunction<Float[]>
, not an IntFunction<float[]>
.
On the other hand, Arrays.stream(wrappers).map(f -> convert1DprimitiveArray(f))
is a Stream<float[]>
so toArray(IntFunction<A[]>)
requires an IntFunction<float[][]>
which is what float[][]::new
is for.
.toArray(float[]::new)
would work if you had a Stream<float>
, which is not possible since you can't have a primitive type as a generic parameter.
Upvotes: 7
Reputation: 15028
Why can I use 'toArray(float[][]::new)' but not 'toArray(float[]::new)
Because after the map
call you have a stream of arrays, and to create an array of the stream, you need to create an array of arrays, not of float
s.
To get the primitive type, you'd have to transform into a DoubleStream
first.
Arrays.stream(wrappers).mapToDouble(FloatWrapper::getFloat)
which gives you a stream of primitives; those you should be able to make float-arrays of with some additional magic.
Upvotes: 2