Added else branches to if constexpr statements to silence MSVC

[vitaut]

Thanks for the PR but could you provide a godbolt repro?

[matt77hias]

It is basically MSVC logging warnings about logically dead code. I'll see if the new compiler version is already available and I can throw something small together.

Edit: I cannot provide a godbolt repro.

• godbolt _MSC_VER == 1939 vs local _MSC_VER == 1940 Also godbolt only supports fmt + MSVC with vcpkg (2023.08.09), not trunk.

I cannot reproduce it on a release build locally. It only happens for debug builds:

...External\fmt\base.h(2713,1): warning C4702: unreachable code
...External\fmt\base.h(2724,1): warning C4702: unreachable code

And then like 100+ of this pair of warnings.

[vitaut]

This is clearly a false positive since the code is reachable for some values of template parameters so I recommend reporting to Microsoft. Because of this and since it's level 4 warning, I suggest using other means to suppress it (e.g. FMT_SYSTEM) unless it can be done in a cleaner way.

[matt77hias]

This is clearly a false positive since the code is reachable for some values of template parameters so I recommend reporting to Microsoft.

Yes, but that is not what MSVC complains about. If the if constexpr condition is true and the if-branch unconditionally returns, the code following the if constexpr statement that is not in an else-branch is dead for that template specialization. It is not complaining for the template specializations for which the if constexpr condition is false.

Imho I would also have thought early-outing from an if constexpr wasn't possible, and one would also need else branches as well. But no idea why MSVC is inconsistent between release and debug builds.