警告：有符号和无符号整数表达式之间的比较 [-Wsign-compare]

Question

Warning in below line

for(int nPort = 0 ; nPort< (sizeof(nArrOverloadParams)/sizeof(int)) && nRetVal 
                    == RET_SUCCESS_VALUE ;nPort++)

Answer 1

The warning indicates that the comparison in principle could fail. The signed value is ⁽¹⁾ promoted to the unsigned type. And if it's negative (the compiler doesn't know that it isn't, without extensive analysis) then the promotion to unsigned produces a very large value.

A consequence is that you're guaranteed that

std::string( "Blah" ).length() < -5

which is just extremely silly – it's a consequence of a choice of type for size values that once made sense, but that that currently is pretty sub-optimal (but impossible to change due to the need for compatibility).

As a rule of thumb it's therefore a good idea to

• use signed types for numbers, but

• use unsigned types for bit level operations.

Eg you can use ptrdiff_t , the type of the result of a pointer difference expression, as a signed version of size_t .

Then you can express a signed size function for arrays as eg

#include <stddef.h>

namespace cppx {
    using Size = ptrdiff_t;

    template< class Item, Size n >
    auto n_items( Item (&)[n] )
        -> Size
    { return n; }
}

and the relevant part of the loop then becomes

using cppx::n_items;

int overloadedParams[77];
for( int port = 0 ; port< n_items( overloadedParams ); port++ )

---

An alternative is to use a range-based loop , bypassing the entire issue:

for( int param : overloadedParams )

If the index is needed for the loop body's logic then a counter can be added.

The behavior of the loop is defined in terms of equivalent code expressed in terms of std::begin and std::end .

---

History. As far as I know it was Dietmar Kuhl who identified the triad of functions begin , end and n_items as practically useful (it's the necessary support for signed size programming), although not with that name for the third function. The C++11 standard added std::begin and std::end , but no std::n_items ( std::extent is a very different beast).

---

^{1) In the standard this conversion is part of the “usual arithmetic conversions” defined by C++14 §5/10.}