signed/unsigned char

signed/unsigned char

[Milena Constantino Caires]

I would like to know why the following statements do not generate a
error during compiling but they do generate segmentation fault and
core dump when the program is running:


char* c = new char(1024);

or

u_char* c = new u_char(1024);

I know that the correct form of this memory allocation is "[1024]" and not
"(1024)" but sometimes it occurs by mistake.

The compiler is not supposed to complain about it???

Thanks in advance,
Milena

Re: signed/unsigned char

[Eljay Love-Jensen]

Hi Milena,

> char* c = new char(1024);

This allocates a single character, initialized with the value 1024.  (Since 1024 is bigger than what a char can hold on most systems, it will get sliced.)

>I know that the correct form of this memory allocation is "[1024]" and not "(1024)" but sometimes it occurs by mistake.

Both forms are correct.

You can allocate an array of char.

You can allocate a single char.

>The compiler is not supposed to complain about it???

No, there is nothing to complain about.  You are allowed to allocate a single object.

HTH,
--Eljay

Re: signed/unsigned char

[random]

Milena Constantino Caires wrote:

>I would like to know why the following statements do not generate a
>error during compiling but they do generate segmentation fault and
>core dump when the program is running:
>
>
>char* c = new char(1024);
>
>or
>
>u_char* c = new u_char(1024);
>
>I know that the correct form of this memory allocation is "[1024]" and not
>"(1024)" but sometimes it occurs by mistake.
>
>The compiler is not supposed to complain about it???
>
>  
>
Because what you have written is that you want a new char whose value is
initalised to 1024. For example write:

To convince yourself this is true, try for example:
#include<iostream>
int main(void) {
  char* c = new char(65);
  std::cout << *c << std::endl;
}

Which should almost certainly print A, unless you are on a very strange
machine (65 is the ASCII value for A).

I'll admit I'm slightly suprised that the compiler doesn't warn you that
1024 is out of range for a char (on the other hand, the code is still
valid I believe if the char is unsigned).

Unfortunatly I don't know of a good way to catch these kinds of errors,
other than of course a tool like valgrind.

Chris