Re: Feature request: gcc should warn upon conversions ptrdiff_t -> int

[Matteo Frigo <athena@fftw.org>]

martin@v.loewis.de (Martin v. Löwis) writes:

> No. In that case, ptrdiff_t most likely would be an integral type with
> 42.69 bits, and conversion to int would cause loss of
> precision. Matteo requests that a warning is issued for size kind of
> conversion (probably also from size_t to int), as on some systems,
> sizeof(ptrdiff_t) > sizeof(int). That would be something like
> -W64bit-portability.

Thanks for clarifying what I was trying to say.

Yes, it would be nice if there were a flag -W64bit-portability that
warns about constructs that behave differently in 32bit and 64bit
machines.  

Besides { ptrdiff_t, size_t, long } -> int conversions, another
potential source of trouble is the following.  Assume

    double *p;       
    unsigned index;  
    int stride;      
    
Consider the expression

     p[index * stride]

If index == 1 and stride == -1, this expression means p[-1] on ILP32
machines, and p[0xFFFFFFFF] on LP64 machines.  It is unlikely that
this was the intended behavior of the code.

(Because int * unsigned is evaluated as unsigned, p[index * stride]
means

     *(p + (ptrdiff_t) ( index *  (unsigned) stride ))

and the conversion unsigned -> ptrdiff_t zero-extends the result to 64
bits.)

Cheers,
Matteo