Re: One-definition rule for inline functions

[Jonathan Wakely <jwakely dot gcc at gmail dot com>]

On 8 March 2012 01:10, Jonathan Wakely wrote:
> On 8 March 2012 01:02, Timothy Madden wrote:
>> Hello
>>
>> I have an inline function like this:
>>
>> void draw_unit_vector()
>> {
>> ? ? ? ?// some large code here
>> ? ? ? ?// ...
>>
>> ? ? ? ?LineTo(1, 1);
>> }
>>
>> in two translation units A and B.
>>
>> Now if the function is large enough it is still compiled and called, and not
>> inlined.
>>
>> The LineTo() function has two overloads:
>> ? ? ? ?- translation unit A only needs the float version:
>> ? ? ? ? ? ? ? ? ?LineTo(float x, float y);
>> ? ? ? ? ?and declares it before it includes the inline function.
>> ? ? ? ?- translation unit B uses both int and float versions:
>> ? ? ? ? ? ? ? ? ?LineTo(int x, int y);
>> ? ? ? ? ? ? ? ? ?LineTo(float x, float y);
>> ? ? ? ? ?and declares them before it includes the inline function.
>>
>> The same inline function would call LineTo(float, float) in source A, but
>> would call LineTo(int, int) in source B. And it so happens that the two
>> overloads for LineTo() are quite different, the int version is for text-mode
>> only, while the float version is for graphics mode.
>>
>> By the C++ standard this case is a clear violation of the one-definition
>> rule, and users should take good care to avoid it.
>>
>> My question is: what does g++ do in such case ?
>>
>> Does it compile two different draw_unit_vector() functions from the same
>> inline function definition ?
>
> Yes, then discards one of the definitions, see
> http://gcc.gnu.org/onlinedocs/gcc/Vague-Linkage.html

Or more precisely, the linker discards one of them.