Zsolt Kiraly said:
> I figured out why the termios program did not work on the Brutus. It was
> not related to termios stuff, rather how chars might not be the same on
> the PC as on the brutus.
>
> In this code there is a function (int foo())that returns an int. But the
> return value is assigned to a char:
>
> char c;
> c = foo();
>
> if (c >= 0)
>
> The arm-linux-gcc (egcs-1.1.1 with your patch + glibc-2.1) warns me that
> this will always be true, due to limited range of the variable. The
> native PC compiler (gcc-2.7.2.3, glibc 2.0.7) does not complain. The PC
> version works flawlessly, the arm version probably assigns 0 to c.
IMHO, this is very bad programming practice. For example, if you're using
getc(), and you're reading characters from a binary file with the following
code:
char c;
do {
c = getc(file);
...
} while (c != EOF);
What happens if the file contains 0xff? The variable that is used
to hold the return value of getc, getopt and so forth should always be
an int. Basically, the problem comes down to a loss of information.
Should you really use a char to old a return value of the size of an
int, and expect to be able to check for values of the int using a char?
> Later in the code:
>
> if (c < 0)
>
> Same warning (will never be true) with the arm gcc, no warning with the
> native gcc.
>
> Workaround:
>
> int temp_c;
>
> temp_c = foo();
>
> c = (char)temp_c; //so I can compare it as a char
>
> if (c >= 0)
This should also be:
if (temp_c >= 0)
> .
> .
> .
> if (temp_c < 0) // because c apperantly does not go < 0
>
> So, I guess the gcc-2.7.2.3 does automatic typecasting, the cross-egcs
> does not ? And the native char seems to be signed, while the cross-char
> unsigned ? I wonder what the ANSI C docs say about this.
The native char is unsigned. It produces better code to have it
unsigned, since you don't have to have two extra instructions after
every load (to convert the unsigned load to a signed value).
> The code is the robin.c program from Erik Troan's Linux Applications
> Development book, and the first warning was line 193, the second at line
> 220.
Most people don't seem to be aware of this issue, because the PC is
what it is.
> After the above modifications the code compiles without errors or
> warnings, and runs on the Brutus (yeee!!!!!!!). Also on the PC.
I have a lot of patches on my FTP site for various programs which suffer
from this problem.
--
Russell King ([EMAIL PROTECTED])
unsubscribe: body of `unsubscribe linux-arm' to [EMAIL PROTECTED]
