By bisecting and quickfixing the autogen issue #18947 (with [autogen.patch](https://gitlab.kitware.com/cmake/cmake/uploads/a222748ec84049de38aceb1400f77ab0/autogen.patch)) I found a performance regression in the CMake run on our main repository (ca 3800 targets in ca 900 files) using Qt on all targets introduced with commit 2f708f5d (~8% slowdown).