-Wimplicit-int-conversion feature request

[steakhal]

PR #126846 started diagnosing unary negations. IMO that makes sense as in the following example it's going to narrow immediately, so the warning seems justified.

However, should we also warn for practically any other unary and binary operators that apply the standard integer promotion to int (if it was shorter than an int before)?
Like in i = i | 1;?

https://godbolt.org/z/3sEEWj8b7

unsigned short unary_negate(unsigned short i) {
    i = ~I; // we get the warning here
    return i;
}

unsigned short binary_or(unsigned short i) {
    i = i | 1; // why not also warn here?
    return i;
}

[steakhal]

@YutongZhuu

[YutongZhuu]

Hi, we introduced that warning because that implicit conversion can lead to incorrect results. For example, assigning

signed_char = -(signed_char)

could be incorrect if signed_char is -128.

In your case, yes, I think C++ internally converts i to an int before doing the binary operation, but I don't think that case would make anything incorrect?

[Explorer09]

I think for the case like i = i | 1, if the result can be converted from signed int to unsigned short without any information loss, the warning won't be worth it.

Because writing in the way to avoid the warning (i.e. i = (unsigned short)(i | 1U)) looks less intuitive and it can mask a potential error in future changes of code due to an explicit cast.

P.S. The warning could be worth it if the bitwise OR mask fits the int type but would be truncated when implicitly cast to unsigned short. E.g. (i | 65536).

[steakhal]

Thank you for the feedback. I agree that my original request does not make much sense.