`ninja all install` does not install _jit_ir_importer binaries/packages

Hi, after building, installing, and exporting libs with:

export PYTHONPATH=:/opt/llvm-project/python_packages/mlir_core:/installs/torch-mlir/python_packages/torch_mlir

I get an error stating that: ModuleNotFoundError: No module named 'torch_mlir._mlir_libs._jit_ir_importer'

If I export the build folder as suggested in the Development.md file, I have no issues.

export PYTHONPATH=:/opt/llvm-project/python_packages/mlir_core:/builds/torch-mlir/tools/torch-mlir/python_packages/torch_mlir

ninja check-torch-mlir-all passes tests (if on green checkmark commit).

One of the missing files that fails to install is here:

/builds/torch-mlir/tools/torch-mlir/python_packages/torch_mlir/torch_mlir/_mlir_libs/_jit_ir_importer.cpython-38-x86_64-linux-gnu.so

How can I install the missing libraries?

My actual project depends on an older version of torch-mlir which does not have published torch-mlir python whees. Directions on generating wheels that include the missing libraries would also be appreciated.

Commands to reproduce:

# clone torch-mlir, initialize submodules and compile with...

mkdir -p /builds/torch-mlir /installs/torch-mlir && \
cmake -GNinja \
    -B/builds/torch-mlir \
    -H/torch_mlir_src/externals/llvm-project/llvm \
    -DCMAKE_INSTALL_PREFIX=/installs/torch-mlir \
    -DCMAKE_BUILD_TYPE=Release \
    -DCMAKE_C_COMPILER_LAUNCHER=ccache \
    -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
    -DLLVM_ENABLE_PROJECTS=mlir \
    -DLLVM_EXTERNAL_PROJECTS="torch-mlir;torch-mlir-dialects" \
    -DLLVM_EXTERNAL_TORCH_MLIR_SOURCE_DIR=/torch_mlir_src \
    -DLLVM_EXTERNAL_TORCH_MLIR_DIALECTS_SOURCE_DIR=/torch_mlir_src/externals/llvm-external-projects/torch-mlir-dialects \
    -DMLIR_ENABLE_BINDINGS_PYTHON=ON \
    -DLLVM_ENABLE_ASSERTIONS=ON \
    -DLLVM_TARGETS_TO_BUILD=host \
    -DTORCH_MLIR_ENABLE_JIT_IR_IMPORTER=1 && \
cd /builds/torch-mlir && \
ninja tools/torch-mlir/all install

Added -DTORCH_MLIR_ENABLE_JIT_IR_IMPORTER=1 just to make sure it is on. It seems to be default to be on.

python -c "import sys; print('\n'.join(sys.path))"

/builds/torch-mlir
/opt/llvm-project/python_packages/mlir_core
/installs/torch-mlir/python_packages/torch_mlir
/usr/lib/python38.zip
/usr/lib/python3.8
/usr/lib/python3.8/lib-dynload
/usr/local/lib/python3.8/dist-packages
/usr/lib/python3/dist-packages
(reverse-i-search)`-c': python ^C "import sys; print('\n'.join(sys.path))"

python -c "import torch_mlir"

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/installs/torch-mlir/python_packages/torch_mlir/torch_mlir/__init__.py", line 18, in <module>
    from torch_mlir.dialects.torch.importer.jit_ir import ClassAnnotator, ImportOptions, ModuleBuilder
  File "/installs/torch-mlir/python_packages/torch_mlir/torch_mlir/dialects/torch/importer/jit_ir/__init__.py", line 14, in <module>
    from ....._mlir_libs._jit_ir_importer import *
ModuleNotFoundError: No module named 'torch_mlir._mlir_libs._jit_ir_importer'

Start with docs/development.md and the ninja tools/torch-mlir/all install command, comparing the build and install package paths shown in the report. Trace how _jit_ir_importer.cpython-38-x86_64-linux-gnu.so is handled, then verify that python -c "import torch_mlir" succeeds from the installed path and that generated wheels include the binary.