const qualifier on function return types.

const qualifier on function return types.

[Lowell Johnson]

Hello,

On IRIX using the MIPSPro C compiler, I get many warnings of the following
sort (GSL 0.7 -- 0.8, at least):

  cc -DHAVE_CONFIG_H -I. -I. -I.. -I.. -O -s -n32 -mips3 -Wl,-woff,85 \
     -fullwarn -c cholesky.c
  ...
  cc-3303 cc: WARNING File = ../gsl/gsl_vector_double.h, Line = 78
    A type qualifier on a return type is meaningless.

    const gsl_vector gsl_vector_const_subvector (const gsl_vector *v,
                                                 size_t i, size_t n);
  ...

I do not get these warnings on Linux/gcc using the following command line:

  gcc -DHAVE_CONFIG_H -I. -I. -I.. -I.. -O -funsigned-char \
      -fwritable-strings -Wall -c cholesky.c

I searched the mailing list archive and saw a short discussion of this and
whether it is ANSI C.  I didn't see any mention of the reason for applying
the const qualifier.

What is the purpose of applying a const qualifier to a return value?  Is it
so that we can see the return value but not change it?

The library builds and works fine, but I would rather not have pages of
warnings in my applications that use the GSL.

Any suggestions?

Thanks.

-- Lowell

Re: const qualifier on function return types.

[Brian Gough]

Lowell Johnson writes:
 > On IRIX using the MIPSPro C compiler, I get many warnings of the following
 > sort (GSL 0.7 -- 0.8, at least):
 > 
 >   cc -DHAVE_CONFIG_H -I. -I. -I.. -I.. -O -s -n32 -mips3 -Wl,-woff,85 \
 >      -fullwarn -c cholesky.c
 >   ...
 >   cc-3303 cc: WARNING File = ../gsl/gsl_vector_double.h, Line = 78
 >     A type qualifier on a return type is meaningless.
 > 
 > The library builds and works fine, but I would rather not have pages of
 > warnings in my applications that use the GSL.

If you have a moment would you mind trying out the program below with
the IRIX compiler.  It's my proposal for a scheme that avoids the
const problem.  It works with GCC but I'd like to hear if it works for
you on IRIX before I go and modify all the view-related functions and
references to them.  Thanks.

Brian


#include <stdio.h>

typedef struct  {
  double * data;
  int size;
} vector ;

typedef struct  {
  vector v;
} view ;

typedef union {
  vector _v;
  const vector v;
} const_view;

void vinc (vector * v)
{
  int i;
  for (i = 0; i < v->size; i++)
    v->data[i] += 1.0;
}

double vsum (const vector * v)
{
  int i; double s=0.0;
  for (i = 0; i < v->size; i++)
    s+= v->data[i] ;
  return s;
}

view mkview (const vector * v)
{
  view w;
  w.v = *v;
  w.v.data++;
  w.v.size = v->size - 1;
  return w;
}

const_view mkcview (const vector * v)
{
  const_view w;
  w._v = *v;
  w._v.data++;
  w._v.size = v->size - 1;
  return w;
}

int
main (void)
{
  double x[] = {123.4, 567.8, 9.123};
  vector v1;

  v1.data = x;
  v1.size = 3;

  printf("v1 = %g .. %g\n", v1.data[0], v1.data[v1.size-1]);

  {  
    const_view cv = mkcview (&v1);
    view v = mkview (&v1);
    printf("v2 = %g .. %g\n", cv.v.data[0], cv.v.data[cv.v.size-1]);
    printf("v2 = %g .. %g\n", v.v.data[0], v.v.data[v.v.size-1]);
    vinc(&v.v);
    vsum(&cv.v);
    printf("v2 = %g .. %g\n", cv.v.data[0], cv.v.data[cv.v.size-1]);
    printf("v2 = %g .. %g\n", v.v.data[0], v.v.data[v.v.size-1]);

  }

  return 0;
}
  

Re: const qualifier on function return types.

[Lowell Johnson]

On Thu, 28 Jun 2001, Brian Gough wrote:
[snip]
> If you have a moment would you mind trying out the program below with
> the IRIX compiler.  It's my proposal for a scheme that avoids the
> const problem.  It works with GCC but I'd like to hear if it works for
> you on IRIX before I go and modify all the view-related functions and
> references to them.  Thanks.
[snip]

Here are the results:

(Copied your test case to a file named gsl_test.c.)

~/tmp> make gsl_test
        cc -O -s -n32 -mips3 -Wl,-woff,85 -fullwarn -o gsl_test gsl_test.c 
~/tmp> ./gsl_test 
v1 = 123.4 .. 9.123
v2 = 567.8 .. 9.123
v2 = 567.8 .. 9.123
v2 = 568.8 .. 10.123
v2 = 568.8 .. 10.123

Let me know if you need additional info.

-- Lowell

Re: const qualifier on function return types.

[Brian Gough]

Lowell Johnson writes:
 > Here are the results:
 > 
 > (Copied your test case to a file named gsl_test.c.)
 > 
 > ~/tmp> make gsl_test
 >         cc -O -s -n32 -mips3 -Wl,-woff,85 -fullwarn -o gsl_test gsl_test.c 
 > ~/tmp> ./gsl_test 
 > v1 = 123.4 .. 9.123


Thanks. No warnings, looks good. I will implement that scheme.  Once
it is done we can release gsl-0.9 which will be the final testing
release.

Re: const qualifier on function return types.

[Brian Gough]

Lowell Johnson writes:
 >   cc-3303 cc: WARNING File = ../gsl/gsl_vector_double.h, Line = 78
 >     A type qualifier on a return type is meaningless.
 >
 > I searched the mailing list archive and saw a short discussion of
 > this and whether it is ANSI C.  I didn't see any mention of the
 > reason for applying the const qualifier.
 >  What is the purpose of applying a const qualifier to a return
 > value?  Is it so that we can see the return value but not change
 > it?

That was the intention, but it doesn't work.  The const on those
return types is meaningless, as the compiler says.

The cause of the warnings is functions like

  const gsl_vector gsl_vector_const_view (const gsl_vector * v, ..)

The logic behind the return type was to preserve constness when
creating vector views, so that a view of a "const gsl_vector *" would
also be const.  It doesn't actually work because the views are
returned as structs and C doesn't preserve constness when copying
structs.

I'll have to do something about it .. it is unsatisfactory right now.