Re: [Mingw-users] How can this floating point behavior be explained?

[JonY <jo...@us...>]

On 8/26/2012 02:21, Dipanjan Das wrote:
> On 25 August 2012 23:44, Keith Marshall <kei...@us...
>> wrote:
> 
>> On 25/08/12 18:54, JonY wrote:
>>> On 8/26/2012 01:37, Dipanjan Das wrote:
>>>> Please consider the snippet below:
>>>>
>>>> *#include <stdio.h>*
>>>> *
>>>> *
>>>> *int main()*
>>>> *{ *
>>>> * printf("%f\n",1);*
>>>> * printf("%f\n",(float)1.0);*
>>>> * printf("%f\n",(int)1.0);*
>>>> * printf("%f\n",1.0);*
>>>> * printf("%f\n",1);*
>>>> *} *
>>>>
>>>> ... snip ...
>>>>
>>>
>>> You are writing code with undefined behavior, bad code is bad code, only
>>> the 2nd printf is valid.
>>
>>
> 
> 
>> Actually, both the 2nd and the 4th are valid; both pass an argument
>> which is promoted to type "double", (a 64-bit entity).  The remaining
>> three pass an argument of type "int", which is only 32 bits, but the
>> "%f" format specifier expects, and consumes 64 bits regardless.  When
>> you pass only 32 bits, the other 32 required could represent anything at
>> all -- hence behaviour is undefined.  JonY is right; your code is wrong,
>> and you can have no expectation of consistent results.
>>
>>
> But, on Turbo C compiler, the first line produces "Floating point format
> not linked" error. I have not found the reason of this behavior though I
> have come across it many a times.
> 

It is part of the reason why your code is undefined, it does not conform
to C standards.

Some compilers may warn, others not, effects are entirely up to the
compiler, OS and hardware combination.