nothing happens here. do we need this dead code and comment then? I suggest to add it when it will be needed and will do something depending on condition.

Those lines worth discussion, I think.

My original intent was to add a guardrail about us refactoring the compiler, but forgetting to define it in headers (or rather defining it in a header other than version.hpp).

However, that attempt failed because apparently we ourselves have an application (sycl-ls) that is built with a 3rd-party compiler (i.e. host gcc or whatever), but uses SYCL headers.

Do we want our headers to be usable with 3rd-party compilers? By that I mean directly using g++ -I/path/to/sycl/includes -include sycl.hpp instead of clang++ --fsycl-host-compiler=g++. We actually have a feature request about this: #5932

If we do want such support, then we should probably define SYCL_LANGUAGE_VERSION macro here. If not, then #error is the right way to go and we need to fix how we build sycl-ls

we do have a warning for this case https://github.com/intel/llvm/blob/sycl/sycl/include/sycl/sycl.hpp#L27
that shows absence of guarantee and doesn't block usage by 3rd party compiler if someone want to try. Why can't we stick to the same approach?

Thanks for reminding me about that diagnostic. If we explicitly claim it to be erroneous, then I would remove the #ifdef completely, providing no guarantee of SYCL_LANGUAGE_VERSION being defined in such mode.