Writing a function that changes an unknown-type numeric variable

[pozz]

In my software, there are some variables that represents some settings.
They usually are numerical variables: unsigned char (0..255),
signed char (-127..128), unsigned int (0..65535), signed int (-32767..32768).

I want to write a generic C function that changes the value of one parameter.
For example, a function that takes a pointer to the variable and increments
it by one. Of course, the function doesn't know the variable type (unsigned
char, signed char, and so on).

I thought to solve this problem in two ways, but I don't like them.

The first solution is to create a struct with 4 different pointers where
only one of that is not NULL.

struct {
unsigned char *p_uchar;
signed char *p_schar;
unsigned int *p_uint;
signed int *p_sint;
...
} param;

But I will lost many RAM space and the incrementing function will be poor
optimized.

if( p_uchar )
*p_uchar++;
else if( p_schar )
*p_schar++;
else if( p_uint )
*p_uint++;
else if( p_sint )
*p_sint++;


The other solution is to use a member of the struct to identify the type of
the variable and using void pointer for the variable.

#define PARAMTYPE_UCHAR 1
#define PARAMTYPE_SCHAR 2
#define PARAMTYPE_UINT 3
#define PARAMTYPE_SINT 4
struct {
int type;
void *p_value;
...
};

In this way I will allocate less space than before but the incrementing function
is low optimized.

if( type==PARAMTYPE_UCHAR )
(*(unsigned char *)p_value)++;
else if( type==PARAMTYPE_SCHAR )
(*(signed char *)p_value)++;
....


I wonder that a different and more elegant solution there is.
I think to the printf() function. I pass int and char with %d parameter
and the printf() works well, but it doesn't know the exact type of
the variable.
Maybe the only problem is with unsigned numerical variables that want
the %u parameter, I think.

 

[Walter Roberson]

I want to write a generic C function that changes the value of one parameter.
For example, a function that takes a pointer to the variable and increments
it by one. Of course, the function doesn't know the variable type (unsigned
char, signed char, and so on).

In this way I will allocate less space than before but the incrementing function
is low optimized.

if( type==PARAMTYPE_UCHAR )
(*(unsigned char *)p_value)++;
else if( type==PARAMTYPE_SCHAR )
(*(signed char *)p_value)++;
...

Try a switch() statement: at worst it will expand in to a chain of
if/else, and the optimizer might be able to do much better than that.

Alternately, try something like,

void increment_uchar( void *p_value ) { *(unsigned char *)p_value)++ };
void increment_schar( void *p_value ) { *(signed char *)p_value)++ };
[...]

Then you have a table of increment functions, and use

increment_functions[type](p_value);

 

[Ben Pfaff]

if( p_uchar )
*p_uchar++;

You mean (*p_uchar)++;

[...]

The other solution is to use a member of the struct to identify the type of
the variable and using void pointer for the variable.

#define PARAMTYPE_UCHAR 1
#define PARAMTYPE_SCHAR 2
#define PARAMTYPE_UINT 3
#define PARAMTYPE_SINT 4
struct {
int type;
void *p_value;
...
};

struct tag {
int type;
union {
unsigned char *uc;
signed char *sc;
unsigned int *ui;
signed int *si;
} p;
};

switch (v->type) {
case PARAMTYPE_UCHAR:
(*v->p.uc)++;
break;
case PARAMTYPE_SCHAR:
(*v->p.sc)++;
break;
case PARAMTYPE_UINT:
(*v->p.ui)++;
break;
case PARAMTYPE_SINT:
(*v->p.si)++;
break;
}

Actually, do you really need pointers? If not, you can do this:

struct tag {
int type;
union {
unsigned char uc;
signed char sc;
unsigned int ui;
signed int si;
} p;
};

switch (v->type) {
case PARAMTYPE_UCHAR:
v->p.uc++;
break;
case PARAMTYPE_SCHAR:
v->p.sc++;
break;
case PARAMTYPE_UINT:
v->p.ui++;
break;
case PARAMTYPE_SINT:
v->p.si++;
break;
}

 

[pozz]

Ben Pfaff ha scritto:

struct tag {
int type;
union {
unsigned char *uc;
signed char *sc;
unsigned int *ui;
signed int *si;
} p;
};

I don't want to use union because I must use const variables that are
stored in a FLASH memory (microcontroller architecture) and not in RAM
that is small. I must initialize that const variables but I have an old
compiler and I can't initialize the members of a union (only C99 can do
that, I think).

Anyway, your example is very similar to the following:

struct {
int type;
void *p_value;
} p;

Or not?

 

[Ben Pfaff]

Ben Pfaff ha scritto:

I don't want to use union because I must use const variables that are
stored in a FLASH memory (microcontroller architecture) and not in RAM
that is small. I must initialize that const variables but I have an old
compiler and I can't initialize the members of a union (only C99 can do
that, I think).

Hmm. Well, that's rough, I guess.

Anyway, your example is very similar to the following:

struct {
int type;
void *p_value;
} p;

It avoids the need for casts or wordy implicit conversions, but
the effect is similar.

 

[Ancient_Hacker]

In my software, there are some variables that represents some settings.
They usually are numerical variables: unsigned char (0..255),
signed char (-127..128), unsigned int (0..65535), signed int (-32767..32768).

I want to write a generic C function that changes the value of one parameter.
For example, a function that takes a pointer to the variable and increments
it by one. Of course, the function doesn't know the variable type (unsigned
char, signed char, and so on).

You can make use of some somewhat fancy macros to do the right thing.
That way there's no long chain of tests at run-time.
Something very close to:

#define DefVar(Name,Type ) typedef Type Name##_Type; Name##_Type
Name;

#define Increment( Name ) (( Name##_Type * ) (&Name) ) ++

 

[Eric Sosman]

Ancient_Hacker wrote On 07/19/06 14:39,:

You can make use of some somewhat fancy macros to do the right thing.
That way there's no long chain of tests at run-time.
Something very close to:

#define DefVar(Name,Type ) typedef Type Name##_Type; Name##_Type
Name;

#define Increment( Name ) (( Name##_Type * ) (&Name) ) ++

If I understand your intent, this would be used as

DefVar(fred,int)
DefVar(ethel,short)
...
Increment(fred);
Increment(ethel);

expanding to

typedef int fred_Type;
fred_Type fred;
typedef short ethel_Type;
ethel_Type ethel;
...
((fred_Type *)(&fred))++;
((ethel_Type *)(&ethel))++;

Have I got it? If so, I don't see how this improves on

int fred;
short ethel;
...
fred++;
ethel++;

(In fact, it has the appearance of a disimprovement.) Would
you mind explaining the benefits, or explaining what I've
overlooked?

 

[Ancient_Hacker]

(In fact, it has the appearance of a disimprovement.) Would
you mind explaining the benefits, or explaining what I've
overlooked?

The original poster wanted a "function" to increment an arbitrary
variable. My humble macro does that, with considerably less run-time
overhead. Whopee.

Although now looking back, it may be that he just has a void * pointer
coming in, in which case this macro is not quite the ticket.

Since the exact original intent isnt clear the best solution is still
up in the air.

Yet another way: If you can segregate the variables by size, then you
can key off the address:

if( Ptr_To_Var >= &Chr_Base && Ptr_To_Var < &Int_Base ) ((uchar *)
Ptr_To_Var) ++
else
if( Ptr_To_Var >= &Int_Base && Ptr_To_Var < &Long_Base ) ((int *)
Ptr_To_Var) ++
else
if( Ptr_To_Var >= &Long_Base && Ptr_To_Var < &End_Vars ) ((long int *)
Ptr_To_Var)
else { // impossible address }

 

[Jack Klein]

In my software, there are some variables that represents some settings.
They usually are numerical variables: unsigned char (0..255),
signed char (-127..128), unsigned int (0..65535), signed int (-32767..32768).

I want to write a generic C function that changes the value of one parameter.
For example, a function that takes a pointer to the variable and increments
it by one. Of course, the function doesn't know the variable type (unsigned
char, signed char, and so on).

I thought to solve this problem in two ways, but I don't like them.

The first solution is to create a struct with 4 different pointers where
only one of that is not NULL.

struct {
unsigned char *p_uchar;
signed char *p_schar;
unsigned int *p_uint;
signed int *p_sint;
...
} param;

But I will lost many RAM space and the incrementing function will be poor
optimized.

if( p_uchar )
*p_uchar++;
else if( p_schar )
*p_schar++;
else if( p_uint )
*p_uint++;
else if( p_sint )
*p_sint++;


The other solution is to use a member of the struct to identify the type of
the variable and using void pointer for the variable.

#define PARAMTYPE_UCHAR 1
#define PARAMTYPE_SCHAR 2
#define PARAMTYPE_UINT 3
#define PARAMTYPE_SINT 4
struct {
int type;
void *p_value;
...
};

In this way I will allocate less space than before but the incrementing function
is low optimized.

if( type==PARAMTYPE_UCHAR )
(*(unsigned char *)p_value)++;
else if( type==PARAMTYPE_SCHAR )
(*(signed char *)p_value)++;
...


I wonder that a different and more elegant solution there is.
I think to the printf() function. I pass int and char with %d parameter
and the printf() works well, but it doesn't know the exact type of
the variable.
Maybe the only problem is with unsigned numerical variables that want
the %u parameter, I think.

The real question is WHY you want to do this. I see from a later post
that you are talking about "const" objects in an embedded system, but
you can't reliably change const objects.

You should explain the real problem that you are trying to solve, and
not just ask for help with what appears to be a rather clumsy solution
you are trying to kludge. There is probably a much better approach to
do what you really need to accomplish.