> On 20 Mar 2025, at 15:57, Thiago Macieira <thiago_at_[hidden]> wrote:
>
> On Thursday, 20 March 2025 02:21:29 Pacific Daylight Time Hans Åberg wrote:
>>> Too late. Warnings turn into errors, so the linker is never run.
>>
>> Why do compiler warnings turn into compiler errors?
>
> Because developers compile their code with -Werror or MSVC's -WX to catch
> warnings in the build and fix them, as most of the warnings that are printed
> are syntactically valid C++, but probably wrong, like type punning or using
> the wrong printf placeholder for a given data type. Fixing the warning does
> not immediately mean you must change the code: you can simply disable the
> warning at that location with a pragma.
>
> But the C++ feature should not rely on developers using pragma, because the
> specific pragmas are not part of the standard. The feature must assume that all
> warnings will be dealt with as "this must be fixed" and developers will just
> change the code.

If they decide compile that way, it is their business, not of the standard. The standard does not specify to issue warnings in the compile, only giving some cases where it might be appropriate.

> That means this feature's only remaining use is that an [[unimplemented]]
> declaration can later be overridden as implemented. You'll need to explain why
> this can't simply be done by changing the first declaration and how that is not
> fragile in the first place. What happens if the declaration order changes?

The same. But if used in a standard library, the order will not change.

>>> [[gnu::weak]] void func();
>>> void f()
>>> {
>>> if (func)
>>> func();
>>> else
>>> fallback();
>>>
>>> }
>>>
>>> With LTO or a sufficiently smart linker, it could detect that "func" is
>>> indeed defined and suppress the check and the else branch.
>>>
>>> But this is not possible in all platforms. So it couldn't be part of the
>>> C++ standard.
>>
>> There are two parts: the warning, and the ability to write a temporary
>> override.
>
> If it's a temporary thing, I can name my function however I want.

Indeed, but the guy who wrote it leaves, and later when the implementation is added, others will have to figure how to fix it instead of doing more important things. A similar question arose on a 600 k lines of C++ code project.