Re: "Arithmetic exception" with modulus operator '%'

Re: "Arithmetic exception" with modulus operator '%'

[Rich Felker]

On Fri, Feb 15, 2019 at 03:35:23AM +0000, jounijl@yahoo.co.uk wrote:
> 
> As in the headline. Program stops and prints "Arithmetic exception"
> at the line where the modulus operator '%' is.
> 
> I'm compiling in Alpine linux with clang installed from apk:s:
> 
> clang -c test.c
> clang -o test test.o
> 
> The code is:
> 
> ----- snip -----
>         unsigned int unum  = 0;
>         unsigned int umod  = 0;
>         unsigned int ures  = 0;
>         ures = unum % umod; // <-- this one
> ----- /snip -----
> 
> The variables have some values other than 0.

I don't follow. You say they have some value other than 0, but the
above example snippet has them as zero, and if they're 0 it's
undefined behavior and a fault of some sort is a typical result. What
did you expect to happen?

Rich

"Arithmetic exception" with modulus operator '%'

[jounijl]

As in the headline. Program stops and prints "Arithmetic exception" at 
the line where the modulus operator '%' is.

I'm compiling in Alpine linux with clang installed from apk:s:

clang -c test.c
clang -o test test.o

The code is:

----- snip -----
         unsigned int unum  = 0;
         unsigned int umod  = 0;
         unsigned int ures  = 0;
         ures = unum % umod; // <-- this one
----- /snip -----

The variables have some values other than 0.

The environment is:
The Alpine Linux is installed in Oracle Virtualbox in FreeBSD 12, 64-bit 
Intel.

$ uname -a
Linux localhost 4.14.89-0-vanilla #1-Alpine SMP Tue Dec 18 16:10:10 UTC 
2018 x86_64 GNU/Linux

$ clang --version
Alpine clang version 5.0.1 (tags/RELEASE_501/final) (based on LLVM 5.0.1)

What does this? Do I need to include some library?

w.b.reg., Jouni

Re: "Arithmetic exception" with modulus operator '%'

[Szabolcs Nagy]

* jounijl@yahoo.co.uk <jounijl@yahoo.co.uk> [2019-02-15 03:35:23 +0000]:
> As in the headline. Program stops and prints "Arithmetic exception" at the
> line where the modulus operator '%' is.
> 
> I'm compiling in Alpine linux with clang installed from apk:s:
> 
> clang -c test.c
> clang -o test test.o
> 
> The code is:
> 
> ----- snip -----
>         unsigned int unum  = 0;
>         unsigned int umod  = 0;
>         unsigned int ures  = 0;
>         ures = unum % umod; // <-- this one

undefined behaviour == anything can happen

the compiler can drop this entirely.

> ----- /snip -----
> 
> The variables have some values other than 0.
> 
> The environment is:
> The Alpine Linux is installed in Oracle Virtualbox in FreeBSD 12, 64-bit
> Intel.

x86_64 traps mod by 0 (in case the compiler didn't
drop the code path because of ub).

> 
> $ uname -a
> Linux localhost 4.14.89-0-vanilla #1-Alpine SMP Tue Dec 18 16:10:10 UTC 2018
> x86_64 GNU/Linux
> 
> $ clang --version
> Alpine clang version 5.0.1 (tags/RELEASE_501/final) (based on LLVM 5.0.1)
> 
> What does this? Do I need to include some library?

what behaviour do you expect?

> 
> w.b.reg., Jouni

Re: "Arithmetic exception" with modulus operator '%'

[jounijl]

Exactly. To be complite:

The host machine prints: "Floating point exception" and outputs a core 
file. Uses: /lib/libc.so.7
The Alpine prints: "Arithmetic exception".  Uses: /lib/ld-musl-x86_64.so.1
Solaris 10 prints: "Arithmetic exception". Uses: /lib/libc.so.1 ; 
/lib/libm.so.2
Ubuntu prints: "Floating point exception" and outputs a core file. Uses: 
/lib/x86_64-linux-gnu/libc.so.6

To the question "what do you except":
Of course the behaviour is similar to others and this is correct. As in 
programs the behaviour would be best like this: number%zero would be the 
number it self when number/zero is undefined or infinity (maby set the 
number to the largest known number). To change this, some mathematical 
evaluation would be needed. Answer: mod 0: Convenient would be the 
number it self ?

Maby the core file has to be enabled somehow. Attached is a complite 
test program. The word "trap" is unfamiliar to me. I think it means just 
checking with an 'if'-like comparison.

Thank you for the replies. It helped. The simple facts are not always 
obvious.

Jouni

the file:

$ cat m.c
#include <stdio.h>
int    main( int argc, char *argv[] );
int    main( int argc, char *argv[] ){
         unsigned int z = 0;
         unsigned int w = 0;
         //floating point exception: fprintf( stderr, "\n 0 %% 0 = %i", 
(int) z%w );
         z = 1;
         //floating point exception: fprintf( stderr, "\n 1 %% 0 = %i", 
(int) z%w );
         z = 0; w = 1;
         fprintf( stderr, "\n 0 %% 1 = %i", (int) z%w );
         z = 1; w = 1;
         fprintf( stderr, "\n 1 %% 1 = %i", (int) z%w );
}

On 14.2.2019 23.58, Rich Felker wrote:
> On Fri, Feb 15, 2019 at 03:35:23AM +0000, jounijl@yahoo.co.uk wrote:
>> As in the headline. Program stops and prints "Arithmetic exception"
>> at the line where the modulus operator '%' is.
>>
>> I'm compiling in Alpine linux with clang installed from apk:s:
>>
>> clang -c test.c
>> clang -o test test.o
>>
>> The code is:
>>
>> ----- snip -----
>>          unsigned int unum  = 0;
>>          unsigned int umod  = 0;
>>          unsigned int ures  = 0;
>>          ures = unum % umod; // <-- this one
>> ----- /snip -----
>>
>> The variables have some values other than 0.
> I don't follow. You say they have some value other than 0, but the
> above example snippet has them as zero, and if they're 0 it's
> undefined behavior and a fault of some sort is a typical result. What
> did you expect to happen?
>
> Rich

Re: "Arithmetic exception" with modulus operator '%'

[Rich Felker]

On Fri, Feb 15, 2019 at 10:31:22AM +0000, jounijl@yahoo.co.uk wrote:
> 
> Exactly. To be complite:
> 
> The host machine prints: "Floating point exception" and outputs a
> core file. Uses: /lib/libc.so.7
> The Alpine prints: "Arithmetic exception".  Uses: /lib/ld-musl-x86_64.so.1
> Solaris 10 prints: "Arithmetic exception". Uses: /lib/libc.so.1 ;
> /lib/libm.so.2
> Ubuntu prints: "Floating point exception" and outputs a core file.
> Uses: /lib/x86_64-linux-gnu/libc.so.6
> 
> To the question "what do you except":
> Of course the behaviour is similar to others and this is correct. As
> in programs the behaviour would be best like this: number%zero would
> be the number it self when number/zero is undefined or infinity
> (maby set the number to the largest known number). To change this,
> some mathematical evaluation would be needed. Answer: mod 0:
> Convenient would be the number it self ?

This has nothing to do with musl or library implementation; what
you're asking for is a *compiler* that defines certain undefined
behavior in a particular way. Even if you had such a thing, writing C
code in order to depend on nonstandard behavior of a particular
compiler would not be a reasonable thing to do.

A better way to achieve the same thing would be just writing a
function that does what you want:

int my_mod(int a, int b)
{
	if (!b) return a;
	else if (b==-1) return 0;
	else return a%b;
}

and using that instead of using the % operator directly. If you need
it to work in constant expression contexts, you could use a macro
instead:

#define MY_MOD(a,b) (!(b) ? (a) : (b)==-1 ? 0 : (a)%(b))

Rich