EXECUTORCH_BUILD_TESTS=ON shouldn't require CMAKE_INSTALL_PREFIX=cmake-out for executorch/schema/program_generated.h

[swolchok]

🐛 Describe the bug

https://github.com/pytorch/executorch/blob/main/test/run_oss_cpp_tests.sh#L31 sets -DCMAKE_INSTALL_PREFIX=cmake-out . I attempted to use cmake manually, because that should let incremental builds work, and since I wasn't installing anything I left -DCMAKE_INSTALL_PREFIX out, but the build failed:

In file included from /Users/swolchok/src/executorch/runtime/executor/test/tensor_parser_test.cpp:9:
/Users/swolchok/src/executorch/../executorch/runtime/executor/tensor_parser.h:15:10: fatal error: 'executorch/schema/program_generated.h' file not found
   15 | #include <executorch/schema/program_generated.h>
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1 error generated.

I believe this reflects a problem with our cmake setup around this generated header.

Versions

Collecting environment information...
PyTorch version: 2.6.0.dev20241218
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: macOS 15.2 (arm64)
GCC version: Could not collect
Clang version: 16.0.0 (clang-1600.0.26.6)
CMake version: version 3.31.2
Libc version: N/A

Python version: 3.12.7 (main, Oct 1 2024, 02:05:46) [Clang 16.0.0 (clang-1600.0.26.3)] (64-bit runtime)
Python platform: macOS-15.2-arm64-arm-64bit
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Apple M1 Pro

Versions of relevant libraries:
[pip3] executorch==0.5.0a0+db8f1eb
[pip3] numpy==1.26.4
[pip3] torch==2.6.0.dev20241218
[pip3] torchao==0.8.0+git2e032c6b
[pip3] torchaudio==2.6.0.dev20241218
[pip3] torchsr==1.0.4
[pip3] torchvision==0.22.0.dev20241218
[conda] Could not collect

[lucylq]

Could you try with #7512 patched?

This seems to be the same issue as #7511

[swolchok]

Could you try with #7512 patched?

Surprisingly, it didn't help. Exact repro command:

cmake -S . -B cmake-out -DCMAKE_BUILD_TYPE=Release -DEXECUTORCH_USE_CPP_CODE_COVERAGE=ON -DEXECUTORCH_BUILD_TESTS=ON -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
                                                                   -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
                                                                   -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
                                                                   -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON \
                                                                   -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
                                                                   -DEXECUTORCH_BUILD_DEVTOOLS=ON \
                                                                   -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_PTHREADPOOL=ON && cmake --build cmake-out -j8

I did rm -rf cmake-out between attempts.

[cptspacemanspiff]

@swolchok
Ok, so I took a very brief look at this...

It seems to be caused by a misconfiguration in the test CMakeList file: executorch/runtime/executor/test/CMakeLists.txt , which assumes that the generated file artifacts are installed before tensor_parser_test can be built.

Specifically, here:

et_cxx_test(
  tensor_parser_test
  SOURCES
  tensor_parser_test.cpp
  EXTRA_LIBS
  portable_ops_lib
  portable_kernels
  extension_data_loader
  extension_runner_util
)
add_dependencies(tensor_parser_test generated_pte_files)
set_property(TEST tensor_parser_test PROPERTY ENVIRONMENT ${test_env})

target_include_directories(
  tensor_parser_test
  PRIVATE "${CMAKE_INSTALL_PREFIX}/schema/include"
          "${EXECUTORCH_ROOT}/third-party/flatbuffers/include"
)

It is including the schemas via the install prefix, which is meaningless before installation.

A quick and dirty change, is to add the library that contains the schema headers program_schema to the extra libs of the et_cxx_test

See below. For what it is worth, I was able to reproduce the error, and this fixed the build, but I am not familiar enough with the test infrastructure to understand what this breaks...

et_cxx_test(
  tensor_parser_test
  SOURCES
  tensor_parser_test.cpp
  EXTRA_LIBS
  portable_ops_lib
  portable_kernels
  extension_data_loader
  extension_runner_util
  program_schema
)
add_dependencies(tensor_parser_test generated_pte_files)
set_property(TEST tensor_parser_test PROPERTY ENVIRONMENT ${test_env})

# target_link_libraries(tensor_parser_test executorch)

# target_include_directories(
#   tensor_parser_test
#   PRIVATE "${CMAKE_INSTALL_PREFIX}/schema/include"
#           "${EXECUTORCH_ROOT}/third-party/flatbuffers/include"
# )

I am honestly confused how this currently builds/and is tested at all, it seems like you would have to build and install the rest of the program first, then rerun the build. This confuses me as I have no idea how these tests work currently in CI.