Can the struct `generic` kindly be renamed (for example to `xgeneric`) to avoid reserved keyword errors with C++/CLI?

[emmenlau]

I've been using the xtensor stack since basically forever. Recently we started incorporating our code into a C# framework that uses C++/CLI to implement native C++ bindings. Everything works well, except of a small problem when xsimd is included: The name generic is a reserved keyword in C++/CLI!

Would it be possible to consider renaming the struct? For example to xgeneric, which would fit xtensors naming convention nicely?

Thanks a lot for the consideration!

[serge-sans-paille]

That's a funny one ^_^
yeah, I don't mind preventing keyword conflict there.

[serge-sans-paille]

@emmenlau would you be able to craft a small reproducer with as few dependencies as possible so that I can include it in CI?

[emmenlau]

Dear @serge-sans-paille , so the thing is, I believe I can create a small reproducer, but in my humble opinion, it would be most sensible to do that in the following way:

• Do not build all of xsimd with C++/CLI, because this is most likely not what users want/need. Usually C++/CLI is used only in an interface layer, where xsimd is consumed.
• Add a single executable, that would build with C++/CLI, and that uses xsimd

Now, the question would be, whether that is very sensible for anything except regression testing in the xsimd CI? I guess most users would really not know why they have such a test executable in their sources, and many may not have a matching compiler to build it - C++/CLI is only supported by cl.exe from MSVC.

For this test to work, CMakeLists.txt needs to tweak compiler flags for this single test, such that "normal" C++ is disabled, and C++/CLI is enabled instead. Its basically removing some compiler flags, and enabling some others. It's not difficult to do, but it may look a bit weired to the non-suspecting user...

Is that then sensible to have?

[serge-sans-paille]

ok. I have no knowledge of C# (/a fortiori/ of C# interfacing with C++) and what I read here doesn't make me want to maintain that. I'll just ask you to double check the change once I made it.

[emmenlau]

Sounds great, thanks a lot! I can already say that I locally patched xsimd 11.2.0 to replace generic with xgeneric and it made the usage of xsimd in C++/CLI work nicely!

[serge-sans-paille]

could generic -> common work too?

[emmenlau]

Yep, I think that would be equally fine!!

[emmenlau]

Just to note, at least in xsimd 11.2, there where quite a number of files named xxx_generic_xxx, and I wanted to avoid renaming files whenever possible. It seemed to me that the name xgeneric is still acceptable with keeping the file names, because they are still moderately similar. But that's really just my two cents...

[serge-sans-paille]

Out of curiosity: why not #define generic xgeneric before including xsimd files and then undef generic in your setup?

[emmenlau]

The main thing is that we use xtensor in hundreds, if not thousands of places in our code base. Now we're starting to generate C# bindings for the code, step by step. Having this workaround for generic is really ok in one or two places, but I fear that for us, it would eventually clutter through all our code. And what's worse, I fear it may lead to funny results if the define is not equally set in all places? Since other people may want to use xtensor with C++/CLI at some point, it would be really much nicer to solve this at the core, if possible...

[serge-sans-paille]

Can you confirm #1114 does the trick?

[emmenlau]

I'll check ASAP!