typecasting of function pointers

[srinivas.satish]

Hi,
is it possible to typecast a function pointer to two different
prototypes.
eg.,
typedef void (functptr1 *) (int , int);
typedef void (functptr2 *) (int);
functptr1 fptr;
fptr = somefunction_name;
fptr(10,20);
fptr = (functptr2)someotherfunction_name;
(functptr2) fptr(10);

I tried the above it did'nt work. If this is wrong please let me know
how it could be done in the right way, also if it is compiler specific,
which compiler(s) permit this.

 

[Michael Mair]

Hi,
is it possible to typecast a function pointer to two different
prototypes.
eg.,
typedef void (functptr1 *) (int , int);
typedef void (functptr2 *) (int);
functptr1 fptr;
fptr = somefunction_name;
fptr(10,20);
fptr = (functptr2)someotherfunction_name;
(functptr2) fptr(10);

I tried the above it did'nt work. If this is wrong please let me know
how it could be done in the right way, also if it is compiler specific,
which compiler(s) permit this.

Please, state exactly what "didn't work" means in future.
Have a look at
<http://www.catb.org/~esr/faqs/smart-questions.html>
And provide the _actual_ code.
The above cannot compile.

So, for the language facilities:
1. Function pointers cannot be converted implicitly (i.e.
without cast). They cannot be explicitly (i.e. via cast)
to another function pointer type in a portable, standard
C way.
This means: Whenever you cast one function pointer type
to another, your program becomes inherently non-portable
and may even break when you use another version of the
same compiler.
2. If you have the same return type, you have good chances
that it will work nonetheless; the right way to do the above
is
void foo(int qux, int quux);
void bar(int baz);

fptr = foo;
fptr(10,20);
fptr = (functptr1) bar;
((functptr2)fptr)(10);
with the typedefs and declarations in place.
However, you can run into deep trouble if calling the wrong
type of function with the wrong number of arguments, so this
is a bad idea at best. If you need different pointer types,
consider using a union of function pointers and some kind of
type marker.
Note: If your parameter list have many parameters (say >5),
then the chance increases that it will not work.
3. On many implementations, even casts between function
pointer types for different return types may work.

Cheers
Michael

 

[Flash Gordon]

Please, state exactly what "didn't work" means in future.
Have a look at
<http://www.catb.org/~esr/faqs/smart-questions.html>
And provide the _actual_ code.
The above cannot compile.

So, for the language facilities:
1. Function pointers cannot be converted implicitly (i.e.
without cast). They cannot be explicitly (i.e. via cast)
to another function pointer type in a portable, standard
C way.
This means: Whenever you cast one function pointer type
to another, your program becomes inherently non-portable
and may even break when you use another version of the
same compiler.

Not completely true. It is perfectly legal to cast from one function
type to another, it's just that you have to cast it back again before
calling it. There are even times when this is useful, although not very
often.

2. If you have the same return type, you have good chances
that it will work nonetheless; the right way to do the above
is
void foo(int qux, int quux);
void bar(int baz);

fptr = foo;
fptr(10,20);
fptr = (functptr1) bar;
((functptr2)fptr)(10);
with the typedefs and declarations in place.
However, you can run into deep trouble if calling the wrong
type of function with the wrong number of arguments, so this
is a bad idea at best. If you need different pointer types,
consider using a union of function pointers and some kind of
type marker.
Note: If your parameter list have many parameters (say >5),
then the chance increases that it will not work.

I would strongly advise against doing this even if it appears to work.
If you do not always use all the parameters then there are two simple
ways to deal with it:
1) Pass suitable 0 values (or some other place marker) for unused
parameters.
2) Write a function that takes a variable number of arguments using the
mechanism C provides for doing this.

3. On many implementations, even casts between function
pointer types for different return types may work.

However, even if it works you should still not do it.

 

[Michael Mair]

Not completely true. It is perfectly legal to cast from one function
type to another, it's just that you have to cast it back again before
calling it. There are even times when this is useful, although not very
often.

Thanks for pointing that out.
For the record: C99, 6.3.2.3#8

I would strongly advise against doing this even if it appears to work.
If you do not always use all the parameters then there are two simple
ways to deal with it:
1) Pass suitable 0 values (or some other place marker) for unused
parameters.
2) Write a function that takes a variable number of arguments using the
mechanism C provides for doing this.

Note: The former may not the natural thing to do; sometimes the
union-based approach is more helpful. The latter, especially
with key/value pairs, does not suffer from that but may make for
troublesome parameter combination checking.

However, even if it works you should still not do it.

Indeed


Cheers
Michael

 

[CBFalconer]

Michael Mair wrote:
.... snip ...


I would strongly advise against doing this even if it appears to
work. If you do not always use all the parameters then there are
two simple ways to deal with it:
1) Pass suitable 0 values (or some other place marker) for
unused parameters.
2) Write a function that takes a variable number of arguments
using the mechanism C provides for doing this.

I advise against 2). Variadic functions do not provide proper
parameter checking, and are the source of many mistakes. They may
be convenient, but can always be replaced by something.

--
"If you want to post a followup via groups.google.com, don't use
the broken "Reply" link at the bottom of the article. Click on
"show options" at the top of the article, then click on the
"Reply" at the bottom of the article headers." - Keith Thompson
More details at: <http://cfaj.freeshell.org/google/>
Also see <http://www.safalra.com/special/googlegroupsreply/>

 

[Ben Pfaff]

I advise against 2). Variadic functions do not provide proper
parameter checking, and are the source of many mistakes. They may
be convenient, but can always be replaced by something.

If you use a compiler that can check variadic function arguments,
then you gain some protection. For example, GCC can check
printf()-style variadic function arguments; newer versions can
also check that an argument list is terminated by a null pointer.
You don't even have to *always* compile with GCC to get some help
from this feature; as long as you sometimes compile with GCC, it
will point out (most of) your mistakes.

 

[Flash Gordon]

Note: The former may not the natural thing to do; sometimes the
union-based approach is more helpful. The latter, especially
with key/value pairs, does not suffer from that but may make for
troublesome parameter combination checking.

I agree there are many other approaches. I was just pointing out a
couple of the simplest to implement which don't rely on doing odd stuff
with function pointers, starting with the very simplest.

<snip more stuff agreed>

 

[CBFalconer]

If you use a compiler that can check variadic function arguments,
then you gain some protection. For example, GCC can check
printf()-style variadic function arguments; newer versions can
also check that an argument list is terminated by a null pointer.
You don't even have to *always* compile with GCC to get some help
from this feature; as long as you sometimes compile with GCC, it
will point out (most of) your mistakes.

That's with a constant format string, and for functions that are
defined in the standard. None of which applies to Flashs suggestion
2.

--
"If you want to post a followup via groups.google.com, don't use
the broken "Reply" link at the bottom of the article. Click on
"show options" at the top of the article, then click on the
"Reply" at the bottom of the article headers." - Keith Thompson
More details at: <http://cfaj.freeshell.org/google/>
Also see <http://www.safalra.com/special/googlegroupsreply/>

 

[Ben Pfaff]

That's with a constant format string, and for functions that are
defined in the standard.

You are wrong about the latter: you can apply these to any
function to you like, using a GCC extension. And, as I said, you
get most of the benefit even if you don't always use GCC.

None of which applies to Flashs suggestion 2.

The null sentinel check might work, or it might not.

 

[Jordan Abel]

That's with a constant format string, and for functions that are
defined in the standard. None of which applies to Flashs suggestion
2.

Or, via __attribute__'s, for a format string returned by a function
guaranteed to return a format string equivalent to its argument, and for
any function which accepts printf-like or scanf-like format strings.

 

[Michael Mair]

... snip ...


I advise against 2). Variadic functions do not provide proper
parameter checking, and are the source of many mistakes. They may
be convenient, but can always be replaced by something.

True; sometimes, this something is so inconvenient that people
have more trouble getting that right or find it too tedious --
especially on extension.
Imagine a function which took a structure until V1.3 -- and then
the facilities have been dramatically extended for V2.0.
Now, initializing 20 of 25 struct members to default for the five
members that matter for the respective application can be very
ugly. The "tag list" concept, i.e. passing key-value pairs was
easier on the eye and made for easier checking than the extended
structure.
As there was no clear distinction between often used and seldom
used members (depending on the users), passing several structures
was quite out of the question, too.
If the whole thing had been designed with the extended feature
set in mind, then there would have been better ways than the
variadic function.
As mentioned in my reply to Flash Gordon: The true pain then
happens internally, if there are no longer consistent default
values and the passed key-value pairs have to be checked for
consistency.

This is a corner case which I have seen at least once; in general,
I agree with the notion to restrict the use of variadic functions
to the ones that come with the standard library (and not all of
them).


Cheers
Michael

 

[srinivas.satish]

Hi,
thanks for all the information

 

[Flash Gordon]

I advise against 2). Variadic functions do not provide proper
parameter checking, and are the source of many mistakes. They may
be convenient, but can always be replaced by something.

I'm no fan of variadic functions either, but they are still better than
casting a function pointer to a type that takes fewer arguments as shown
in the argument that started this thread.

My personal preference is for keeping the number of arguments small and
using suggestion 1, which is why it was my first suggestion.