[Build] Fix `FetchContent` multiple build issue (#9596)

Signed-off-by: luka <luka@neuralmagic.com>

CMakeLists.txt

@@ -169,12 +169,12 @@ endif()
 
 #
 # Use FetchContent for C++ dependencies that are compiled as part of vLLM's build process.
-# Configure it to place files in vllm/.deps, in order to play nicely with sccache.
+# setup.py will override FETCHCONTENT_BASE_DIR to play nicely with sccache.
+# Each dependency that produces build artifacts should override its BINARY_DIR to avoid
+# conflicts between build types. It should instead be set to ${CMAKE_BINARY_DIR}/<dependency>.
 #
 include(FetchContent)
-get_filename_component(PROJECT_ROOT_DIR "${CMAKE_CURRENT_SOURCE_DIR}" ABSOLUTE)
-file(MAKE_DIRECTORY "${FETCHCONTENT_BASE_DIR}")
-set(FETCHCONTENT_BASE_DIR "${PROJECT_ROOT_DIR}/.deps")
+file(MAKE_DIRECTORY ${FETCHCONTENT_BASE_DIR}) # Ensure the directory exists
 message(STATUS "FetchContent base directory: ${FETCHCONTENT_BASE_DIR}")
 
 #
@@ -509,6 +509,8 @@ else()
           GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
           GIT_TAG 013f0c4fc47e6574060879d9734c1df8c5c273bd
           GIT_PROGRESS TRUE
+          # Don't share the vllm-flash-attn build between build types
+          BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn
   )
 endif()
 

setup.py

@@ -157,6 +157,14 @@ class cmake_build_ext(build_ext):
         # on subsequent calls to python.
         cmake_args += ['-DVLLM_PYTHON_PATH={}'.format(":".join(sys.path))]
 
+        # Override the base directory for FetchContent downloads to $ROOT/.deps
+        # This allows sharing dependencies between profiles,
+        # and plays more nicely with sccache.
+        # To override this, set the FETCHCONTENT_BASE_DIR environment variable.
+        fc_base_dir = os.path.join(ROOT_DIR, ".deps")
+        fc_base_dir = os.environ.get("FETCHCONTENT_BASE_DIR", fc_base_dir)
+        cmake_args += ['-DFETCHCONTENT_BASE_DIR={}'.format(fc_base_dir)]
+
         #
         # Setup parallelism and build tool
         #