On 8/14/18, Joseph Myers <jos...@codesourcery.com> wrote:
> On Tue, 14 Aug 2018, Martin Jambor wrote:
>
>> when you try compiling a call to function abs and provide an unsigned
>> int in the argument in C++, you will get an error about ambiguous
>> overload. In C however, it will pass without silently. The following
>> patch adds a warning for these cases, because I think it is likely that
>> such code does not do what the author intended.
>
> abs of unsigned short (promoted to int) seems harmless; I don't see any
> tests to make sure it doesn't warn. Really the issue seems more like abs
> (or labs / llabs / imaxabs) of an argument whose value might be changed by
> the conversion to int. Except that's more like a subset of -Wconversion,
> and highly likely to have false positives (for a long argument that is
> actually always in the range of int), so maybe a restriction to unsigned
> arguments makes sense for this warning, but should still only be for
> unsigned arguments at least as wide as the argument type of the abs
> function in question.
>
>> + else if (DECL_FUNCTION_CODE (expr.value) == BUILT_IN_ABS
>
> This looks like it would only handle abs, not labs / llabs / imaxabs.
>
>> +@code{<} or @code{>=}. When compiling C, also warn when calculating
>> +an absolute value from an unsigned type. This warning is also enabled
>
> But this would suggest any absolute value function, not just abs.
clang has a -Wabsolute-value warning flag that might be looking at
here for comparison; see bug 63886 for an example:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63886
>
> --
> Joseph S. Myers
> jos...@codesourcery.com
>
