Eclipse indexer does not follows supplied additional compilation flags for Cuda toolchain

Hello all,
we develop CUDA app in Eclipse CDT Version: 2022-09 (4.25.0) with Nsight plugin which is part of CUDA 12.1 sdk.
We build the project with our own makefiles and it works fine. But there are some troubles with eclipse indexer where compiler options/setting diverges from that in makefiles. Particularly c++ dialect setting which causes false errors in IDE and prevents from correct code hints. I tried to follow some advices from Eclipse and Stack Overflow forum and tried to unify them by adding required flags directly to compiler invocation command for indexer purpose. See picture below.

This setting works correctly when I change indexer toolchain to GCC but not for CUDA toolchain. But we do need using cuda toolchain to make indexer working for *.cu files. I also tried to execute the command in the picture directly from external linux terminal. $(COMMAND) was substituted by nvcc. And it surprisingly works as expected and takes into account added dialect option. So I guess a problem could be somewhere in Nsight plugin and what command it uses to invoke compiler for indexer purpose.

Does anyone have similar experience? Any help or suggestions how to solve this are welcomed.

Regards Ondrej