Hello everybody,
I am starting this thread because I encounter a problem with the pointers in Fortran.
I want to use a pointer of an array-valued function, but I get a segmentation fault error during execution (compilation is ok with gfortran).

Everything works well when the pointer aims at a function returning a real. But it fails when the function returns an array of reals. I guess there is a problem in the definition of my pointer, but I don't know what to do.
Here is a small code to illustrate my problem :


Any help would be appreciated.
Thanks.

Two suggestions.

First, I don't think you can use the same pointer to point to the two functions...if one of them is an array, I would suggest to create another pointer to point to f2, like this:

real*8, pointer :: ap(:)

I called it 'ap' for 'array pointer'

If that alone does not work, I wonder if you should turn your f2 into a pointer itself and THEN return it:

function f2(x)
implicit none
real*8, pointer :: f2(:)
real*8, intent(in) :: x
allocate(f2(2))
f2(1) = x+1
f2(2) = x-1
end function f2

As far as I know - but I cannot check now cause I am waz from home - you cannot use pointers to point to functions, at least f90 / 95, you can have them point at variables only (do not know about constants).

Any pointer in fortran is dereferenced at once. You decalred your pointer to point at a variable of typa real*8. A pointer to an array must be delared as array like

real*8, dimension (:), pointer :: ap



Norbert

The optimist believes we live in the best of all possible worlds - the pessimist fears this might be true.

Don't know why you want to use pointers on functions - but maybe it's only, because you want to use several functions in the same computation, like this:
p => f1
result = compute_something(p)
p => f2
res = compute_something(p)

In this case you can rather use functions as arguments of other functions, i.e.:
result = compute_something(f1)
result = compute_something(f2)

Thank you for your answers.
Here are my comments to all of your recommendations.

to Salgerman (post #1):
The problem is that I do not to point at a variable, but at a function. Therefore, I cannot declare my pointer with
What I tried to do is to use the following declaration :
but in that case I get the following error during the compilation :
However, here I do not want to declare an array of pointers, but a pointer that points to an array-valued function, which is different.
As a consequence, your second option does not work neither.

to gummibaer :
pointer of functions have been introduced in Fortran 2003. It works well when the functions are returning a single value, but I have some problems when an array is returned.

I finally found how to do this, using 2 different pointers (p and ap) and an explicit interface. Here is my code :


to Salgerman (post #2):
You are right, this is exactly what I want to do, but I don't know how to declare the type of the function f passed as argument in the function 'compute_something', that returns a real*8 for f1 but an array of size 2 of real*8 for f2. I initially thought it would be easier using pointers of functions, but it appears it is not.

Here is an illustration of what I want to do, and the problem I get when calling the subroutine 'compute_something' with 2 different functions types as argument (f1 returns a real*8, f2 returns an array of real*8) :

the compilation error for the line "call compute_something(f2)" is :


ps : I used two modules, m1 and m2, because I get a compilation error when including m1 in the contains part of the main program. I don't undertand why, anyway...

Here is an working example for a function which has an argument scalar function:
http://www.tek-tips.com/viewthread.cfm?qid=1570169

If you want to use as function argument an function which returns an array, you need to use interface. For an working example see funcargs.f95 in the last posting with in this thread:
http://www.tek-tips.com/viewthread.cfm?qid=1572869

You links seem to contain interesting information.
I will look at them with great attention and come back later to tell if it works for my case.
Thank you !

I think you need something more special. Maybe you want one function which will operate with both types of arguments: scalars and vectors.

Simply said, I want only one function my_function with 2 arguments and I want to apply the function on scalar arguments (real function and real number) and on vector arguments (vector function and vector vector). This is possible in fortran too - we can implement function overloading suing generic interface.
Here is an example
overloading.f95

You see, I need write the function my_function_with_scal_args with scalar arguments and the function my_function_with_vec_args with vector arguments.
Then implement the overloading mechanism via generic interface my_function.

Now it compiles and runs:

Thank you, this is exactly what I was looking for ! I tried with my own example and it works very well.
One last question : to avoid the use of the function overloading with the general interface, I was wondering if it was possible to define, in the interface for the function passed as argument, the arguments as assumed-shape arrays. In that way, I define only once my function, that combines both 'my_function_with_scal_args' and 'my_function_with_vec_args', by writing the interface in a way like this one :

With my (simplified) previous code, it writes :

unfortunately, if it compiles without problem, I got a segmentation fault error when the line 'print*,f(1.0D0)' in the subroutine 'compute_something' is executed.
Is it possible to do such a thing ? It would be very interesting for me because, with such an approach, I can write a single function that computes, for instance, the Jacobian matrix (size m*n) of a vectorial function for any values of m and n. By using the function overloading and the general interface, I would have to write the m*n functions !

I cannot simply answer your questions, it would be best when you try it self.
For example look at this thread
http://www.tek-tips.com/viewthread.cfm?qid=1546213
The function multiply_permutations has 2 arguments, i.e. arrays without defined dimension and returns allocatable array...

Here is other example
http://www.tek-tips.com/viewthread.cfm?qid=1516435
Look at the function MatrixVector(M, x) in module vector_functions. It has 2 input arguments

real, dimension(:,:), intent(in) :: M
real, dimension(:), intent(in) :: x

and returns a vector

real, dimension(n) :: MatrixVector

which size depends on input argument / n=SIZE(x) /

Thank you for your help. I am going to look at those links and post my answer with, I hope, the solution as soon as possible.

I had a look on the links you provided, mikrom, but unfortunately they deal with with simple assumed-shape arrays, and not assumed-shape arrays of functions, and are helpless for my case.

Let me sum up my problem, for the readers of the forum.
I have the code given below. In this code, the function f returns an array (dimension 2) of reals. This function is passed as argument to the subroutine mysub.
What I would like to do is to be able to pass any function that returns an array of dimension n >= 1 to mysub. I tried modifying the line
to
in the interface inside mysub, to act as an assumed-shape array, but I get a segmentation fault during execution (using gfortran or f95).
Does anyone know the solution of my problem ? I searched for a long time on the Internet but didn't find any similar thing.

I don't quite understand if what you are trying to do is to take on any one-dimensional function of any length or any-dimensional.

The code below works for me and uses mysub to take on 2 one-dimensional functions of different lengths.

typo!!! Sorry....need to change "allocate(f5(2))" to "allocate(f5(5))" !!!

and the indices (2) (2) (2) to (3) (4) (5) ...

What you want is possible.

I modified a little bit the program:
Now the function vf1() takes a vector argument of any dimebnsion and returns the vector of the same dimesion as the input argument.
The function my_function_with_vec_args() accepts now arguments like vf1()

overloading2.f95

The program a vector of length 2 and 5, the function vf1() returns in that cases a vector of length 2 or 5.

But I must say you, that what I have done is compiler dependent. It compiles with gfortran, but it doesn't compile with g95, which doesn't like this declaration:
and throws this error:

That you for your answer, both showing what I want to do with different methods.

salgerman, you understood my problem welle, I wanted to take on any one-dimensional function. I am just surprised I get a segmentation fault if I do not declare f2 and f5 as allocatables. Anyway, by declaring them as allocatable it works fine and that is what I will do in the future.

mikrom, I am very impressed by your solution, where it not not necessary do declare the functions with allocatables. I need a bit more time to understand clearly what it does exactly. I tried to reproduce my example with your solution but encountered some errors. I guess I need to reread your code carefully to fully understand it.

anyway, thanks very much to both of you for your help. It is greatly appreciated.
Best regards.