C4702: unreachable code

[hanusek]

Hello!
I have a problem:

C:\builds\eprotect-platfrom\auth\client\include\nonstd\expected.hpp(2142) : error C2220: the following warning is treated as an error
C:\builds\eprotect-platfrom\auth\client\include\nonstd\expected.hpp(2142) : warning C4702: unreachable code

OS: Windows 2022
Visual Studio 2022 v17.5.0

https://conan.io/center/recipes/expected-lite?version=0.6.3

[subbota-a]

Me too

[szaszm]

How does that work? That line is a comment, there is no code there: https://github.com/martinmoene/expected-lite/blob/v0.6.3/include/nonstd/expected.hpp#L2142
Is it any different in your build directory?

[np-c0mp1ete]

Hello!

Repro:

#include "expected.hpp"

int main()
{
    nonstd::expected<int, int> ei = 0;
    return ei.value();
}

Compile using cl /O2 /W4 /WX main.cpp. expected-lite version is 0.6.3.

Output:

Microsoft (R) C/C++ Optimizing Compiler Version 19.39.33523 for x64
Copyright (C) Microsoft Corporation.  All rights reserved.

main.cpp
C:\test\expected.hpp(1803) : error C2220: the following warning is treated as an error
C:\test\expected.hpp(1803) : warning C4702: unreachable code

[martinmoene]

On Compiler Explorer:

• MSVC v19.latest: -O2 -W4 -WX -Dnsel_CONFIG_SELECT_EXPECTED=nsel_EXPECTED_NONSTD: C4702: unreachable code.
• clang 18.1.0: -O2 -Wall -Wunreachable-code-aggressive -Dnsel_CONFIG_SELECT_EXPECTED=nsel_EXPECTED_NONSTD: no warning.

I'm not sure what to think of MSVC's warning.

[szaszm]

Ah, at first I thought line 2142 is with the 0.6.3 release branch, but apparently it's on master, so there is actual code there. So the correct reference is

• https://github.com/martinmoene/expected-lite/blob/v0.6.3/include/nonstd/expected.hpp#L1803
• https://github.com/martinmoene/expected-lite/blob/master/include/nonstd/expected.hpp#L2142

Maybe MSVC doesn't like the value() call next to the comma operator after rethrow, which is actually unreachable. Maybe a random throw expression instead of value() call would silence it, but I can't test this now.

A few other wrong ideas I had:

• immediately invoked lambda that calls rethrow, and lies about its return type. That would produce another warning about the function not returning anything, because rethrow is not actually throwing.
• changing rethrow to lie that its return type is value_type&. That would generate a similar warning inside rethrow, because it doesn't return that type, nor does it throw. assert and RaiseException look like normal function calls to the control flow analyzer.
• change to an if statement. This wouldn't work with C++11 compilers in constexpr context.

[martinmoene]

Thanks for your helpful thoughts to advance this issue.