( C ) How do you declare a function pointer where a parameter is a function pointer?

spursrule · 11-25-2007, 05:25 PM

I'm writing a simple program to time various sorts, which can be configured by command-line switches.

Assume I have a family of sorts with prototypes of the same format as:

Code:

int insertion_sort(void *A, int n, int sz_elem,
      int (*cmp)(const void *p, const void *q))

i.e., like Strousstrup does on pg 158 of "The C++ Programming Language" in his shellsort.

Where cmp is a comparison predicate and sz_elem is the size of any one element (all the same size) in the array A.

What I want to do is have a function pointer called sort, so I can assign the correct function to sort, and just call

Code:

(*sort)(A,n,sizeof(int),some_cmp)

if say, A is an array of ints.

My first thought was to declare sort as:

Code:

int (*sort)(void*, int, int,
          int (*)(const void*, const void *));

However, gcc told me no. I know it can be done pretty easily as

Code:

typedef  int (*Cmp_fcn)(const void*, const void*);
int (*sort) (void*, int, int, Cmp_fcn);

However, how could I do it without the typedef?

Code:

int (*sort) (void*, int, int, int);

gives a warning about assignment from incompatible pointer type, so it can't be right (right?).

Please don't laugh at me for looking in a C++ book for a way to write in C-style instead of C++-style.
lol

Thanks

ta0kira · 11-25-2007, 05:43 PM

At what point did gcc say "no"? Was it the declaration of sort, or its assignment? I don't see a reason to not use a typedef, or a reason why your non-typedef declaration shouldn't work. It's a hell of a lot less confusing to typedef for someone who needs to read your code and it gives you more time to figure out real problems

ta0kira

PS Comeau likes both versions, so it could be a gcc bug. Or it could be that your problem is with a cast?

spursrule · 11-25-2007, 06:04 PM

Quote:

Originally Posted by ta0kira

At what point did gcc say "no"? Was it the declaration of sort, or its assignment? I don't see a reason to not use a typedef, or a reason why your non-typedef declaration shouldn't work. It's a hell of a lot less confusing to typedef for someone who needs to read your code and it gives you more time to figure out real problems

ta0kira

PS Comeau likes both versions, so it could be a gcc bug. Or it could be that your problem is with a cast?

Oh man, chalk this up to programmer error. I must have frustratingly typed something incorrectly the first time, because now it works fine with the exact un-typedef-ed form of the declaration from above. I was asking because I felt like I might have been missing some big-picture thing with function pointers if I couldn't prototype/declare a function that had a function pointer as a parameter. I must have forgot a * somewhere or something. I completely agree on using the typedef-ed version to improve (or maybe give?) readability.

Thanks though.

jim mcnamara · 11-26-2007, 04:21 PM

my gcc 4.1 likes this:

Code:

#include <stdlib.h>
typedef
void (*ss_t) (void *,
               size_t,
               size_t,
               int (*cmp)(const void *, const void *));
void shell_sort(void *, size_t, size_t, int (*cmp)(const void *, const void *));
void ins_sort(void *, size_t, size_t, int (*cmp)(const void *, const void *));

ss_t sorts[3]={shell_sort,
               ins_sort,
               qsort};

bigearsbilly · 11-27-2007, 04:37 AM

I would ddeclare a function pointer type as a typedef first
like the X libs do, as in Intrinsic.h

Code:

typedef void (*XtActionProc)(

    Widget              /* widget */,
    XEvent*             /* event */,
    String*             /* params */,
    Cardinal*           /* num_params */

);


... later on ....
extern void XtRegisterGrabAction(

    XtActionProc        /* action_proc */,
    _XtBoolean          /* owner_events */,
    unsigned int        /* event_mask */,
    int                 /* pointer_mode */,
    int                 /* keyboard_mode */

);

It makes it all a bit more readable, n'est-ce pas?

ta0kira · 11-27-2007, 07:56 PM

Quote:

Originally Posted by spursrule

I completely agree on using the typedef-ed version to improve (or maybe give?) readability.

In one of my projects, I have a header with about 30 typedefs mostly meaning the same things like void*, uint32_t, etc. It tells people what you intend an argument or variable to mean and it also makes maintenance a lot easier. For example, I realized that I needed to change a certain type of argument from unsigned to signed in a 150+ file project and all I had to do was change one line in the typedef file.
ta0kira