noexceptions

[20162026]

add support for builds with exceptions disabled #72

[20162026]

• We can add the BEMAN_INPLACE_VECTOR_NO_EXCEPTIONS CMake option and generate config.hpp, but I would like to keep defined(__cpp_exceptions) as it's the industry-standard way to handle noexception build.

• current test implementation is ugly, but I dont know how to improve it while keeping gtest

[wusatosi]

Thanks for the contribution as always, I will review tomorrow.

[wusatosi]

See comments

[wusatosi]

We can add the BEMAN_INPLACE_VECTOR_NO_EXCEPTIONS CMake option and generate config.hpp, but I would like to keep defined(__cpp_exceptions) as it's the industry-standard way to handle noexception build.

This does make sense to me, I don't know if we want to go with this, we have a meeting agenda on discussion on the config.hpp approach of doing things vs industry standard direct macro detection, I will bring this up.

[wusatosi]

Hey @20162026 , we will bring up some feedback against CPP.NO_FLAG_FORKING at the weekly sync the coming Monday, I think you have some valuable experience interacting and developing with this rule, would you mind writing a short feedback here so I can forward it to our meeting agenda?

[20162026]

@wusatosi some arguments I could think of right now, mostly regarding this and header only libs in general:

1. I don't think that ODR violation prevention measures should be enforced for header only libraries, as it is up to the user how the compilation units that are using the library should behave.
2. For header only (especially single header) libraries, it is inconvenient to have cmake generated config files, as the end user often just wants to drag and drop the headers or use a different build system, and would be forced to manually generate the config file
3. This is more of a unique case and can most likely be avoided by splitting the freestanding impl into a different namespace. However, I have a project where I would like to use both HOSTED and FREESTANDING-DELETED versions of inplace_vector. If we used a preprocessor define to distinguish between them and enforce it at top level, it would require multiple instances of the same library.
4. From personal observation, it is common practice for all of the standard libraries to FLAG FORK

[wusatosi]

Thanks for the comments, will forward.

[wusatosi]

oh also, how do we refer to you? : )

[20162026]

@wusatosi my name is Edgar if that's what you are asking.

Also, PR should be more or less ready, besides the NO_FLAG_FORKING compliance

[wusatosi]

We can add the BEMAN_INPLACE_VECTOR_NO_EXCEPTIONS CMake option and generate config.hpp, but I would like to keep defined(__cpp_exceptions) as it's the industry-standard way to handle noexception build.

Do you think the CMake option that has its default value based on if __cpp_exceptions is working would be acceptable?

I can set this up directly on this PR, would you mind that?

[20162026]

Do you think the CMake option that has its default value based on if __cpp_exceptions is working would be acceptable?

@wusatosi do you mean checking it in the config.hpp or in CMake itself using something like check_cxx_source_compiles ? I have pushed an example of how I would imagine it.

I can set this up directly on this PR, would you mind that?

Feel free to push any changes you see fit.
PS. there's no need to ask, I’d just uncheck the Allow edits by maintainers flag on GitHub otherwise

[wusatosi]

Feel free to push any changes you see fit.
PS. there's no need to ask, I’d just uncheck the Allow edits by maintainers flag on GitHub otherwise

Noted, I usually find it rude for people to push directly on my branch, so I always ask HAHA.

[wusatosi]

Thank you so much for all the work!

I have one minor question but besides that everything looks great.

[wusatosi]

Is this still needed?

[20162026]

Yes, it ensures that there are no throws in the compiled code. Without it, EXPECT_DEATH could still pass if an unhandled exception was thrown.

Although, I'm now not sure if we need the compiler check. It allows running the tests on compilers that don't support or have a different syntax for disabling exceptions (e.g. MSVC). But right now, CI only includes Clang and GCC, and maybe it would be better to error out instead of silently skipping tests on other compilers. On the other hand, it doesn't matter much, as the implementation should be compiler agnostic as long as there is abort()

        CMAKE_CXX_COMPILER_ID STREQUAL "Clang"
        OR CMAKE_CXX_COMPILER_ID STREQUAL "GNU"