[clang] [clang][Darwin] Remove legacy framework search path logic in the frontend (PR #120149)

[Joseph Huber via cfe-commits]

================
@@ -1,13 +1,6 @@
-// RUN: %clang -cc1 -fcuda-is-device -isysroot /var/empty \
-// RUN:   -triple nvptx-nvidia-cuda -aux-triple i386-apple-macosx \
-// RUN:   -E -fcuda-is-device -v -o /dev/null -x cuda %s 2>&1 | FileCheck %s
-
-// RUN: %clang -cc1 -isysroot /var/empty \
-// RUN:   -triple i386-apple-macosx -aux-triple nvptx-nvidia-cuda \
-// RUN:   -E -fcuda-is-device -v -o /dev/null -x cuda %s 2>&1 | FileCheck %s
-
 // Check that when we do CUDA host and device compiles on MacOS, we check for
 // includes in /System/Library/Frameworks and /Library/Frameworks.
 
-// CHECK-DAG: ignoring nonexistent directory "/var/empty/System/Library/Frameworks"
-// CHECK-DAG: ignoring nonexistent directory "/var/empty/Library/Frameworks"
+// RUN: %clang -isysroot /var/empty -target unknown-nvidia-cuda -v -fsyntax-only -x cuda %s -### 2>&1 | FileCheck %s
----------------
jhuber6 wrote:

When we detect CUDA mode we create a CUDAToolChain which is a separate offloading ToolChain that handles the device-side compile. That contains the `HostTC` which should be Darwin in this case, the calls to get the include headers should then be routed through the CUDA ToolChain to that one I believe.

https://github.com/llvm/llvm-project/pull/120149