Re: [CIL users] Weird behavior with functions used before being declared

[Christoph S. <cs...@fr...>]

Elnatan -

On Fri, Oct 31, 2008 at 01:19:34PM -0400, Elnatan Reisner wrote:
> CIL adds prototypes for functions that are called before they are
> declared, but the transformations seem to treat the return value as an
> integer regardless of the real return type. For example, the following
> program prints '0' when run through CIL. This is because CIL passes the
> value f(x) through a temporary *integer* variable, so the value is
> truncated.
>
> #include <stdio.h>
> int main() {
>         long long x = 0x100000001LL;
>         printf("%d\n",f(x) == x);
>         return 0;
> }
> long long f(long long x) {
>         return x;
> }
>
> Now, that program doesn't compile with gcc because of conflicting types
> for f (because the implicit declaration says it returns an int), so you
> could say that all bets are off.

Brilliant conclusion, Dr Watson!

1. Your example program is illegal C: In main()
function f() is called without a prototype.  The
Standard says that every argument and the return
type of such a function is "int".  The
definition of f (which includes a prototype)
later on contradicts the declaration implied by
the call earlier on.

2. Gcc is perfectly right in refusing this
program.

3. You found a bug in Cil because it accepts bad
code without an error message.


> But I would think that the two reasonable options would be:
> 1. Have CIL complain (like gcc does), rather than compile such code.
> This means, I think, that missing prototypes should only be added for
> functions with a return type of int.

IMHO this is the correct way to go.  Cil should
flag the error and not generate an output file.

Do You have a patch against the current svn that
does accomplish this?


> 2. Leave things as the types they are.
> Compiling the program but casting back and forth from int seems to me
> like an odd choice. Do others agree?

No.  We must never accept illegal code.


/Chris

--
Dr. Christoph L. Spiel