-Wunused-lambda-capture too strict with variadic usage

[JVApen]

Bugzilla Link	41211
Version	8.0
OS	Windows NT
CC	@zmodem,@JVApen,@pepsiman,@zygoloid,@Weverything

Extended Description

Reproduction on compiler explorer: https://godbolt.org/z/hJHQui
error: lambda capture 'i' is not used [-Werror,-Wunused-lambda-capture]

In the example below, we get a compiler warning on 'unused' lambda capture 'i'.
when changing the call to f from 'f<>' to 'f', the warning goes away.
When removing the capture, as suggested, we get: error: variable 'i' cannot be implicitly captured in a lambda with no capture-default specified

t.cpp

template <typename ...T>
int h(T &&...t);

template
int g(int i);

template<typename ...T>
auto f(int i)
{
return i{ return h(g(i)...); };
}

int main()
{
return f<>(42)();
}

run.sh

clang++ -std=c++17 -O3 -Wunused-lambda-capture -Werror t.cpp

[JVApen]

Correcting 'enhancement' to 'normal' issue. This ain't a feature request.

[zmodem]

rtrieu, as a warnings expert, do you have any thoughts here?

[Weverything]

I think what's happening is that f<>() is instantiating the lambda to:

return i{ return h(); };

and f() is instantiating it to:

return i{ return h(g(i)); };

So technically, in the instantiation, i is unused, but in the template pattern, it is used. The solution is to remove the warning in template instantiations and just warn in the template pattern. That should remove the erroneous warning.

[zmodem]

Richard, would you be interested in preparing a patch?

Or Malcolm perhaps?

Regardless, I don't think we can block the release on this.

[Weverything]

I just checked what would happen if the warning was turned off the template specializations and turned on only for template patterns. Unfortunately, it looks like there is a difference between how the ODR-used tracking is computed between the two modes, so several false positive warnings pop up.

[pepsiman]

Can it be turned off for instantiations of variadic templates with an empty parameter pack?

[zmodem]

This does not seem to be getting fixed for clang 10, and I don't think we can block on it.

[JVApen]

Is there any update on when this can get fixed?

[llvmbot]

@llvm/issue-subscribers-c-11