MLIR Python bindings: .pyi imports?

MLIR
,
5

Hi all, sorry to re-open this thread,
but I’m facing the same issue when trying to use an LLVM build that was built (with the MLIR Python bindings) on a different machine.
This is how I import the bindings in my script:

sys.path.insert(0, "/data/llvm_build_fffdd9ed/python_packages/mlir_core")
from mlir.ir import Context, Module, Operation, Region, Block, Value

and this is the error that I’m getting:

File "../myScript.py", line 15, in <module>
    from mlir.ir import Context, Module, Operation, Region, Block, Value
  File "/data/llvm_build_fffdd9ed/llvm/python_packages/mlir_core/mlir/ir.py", line 5, in <module>
    from ._mlir_libs._mlir.ir import *
  File "/data/llvm_build_fffdd9ed/llvm/python_packages/mlir_core/mlir/_mlir_libs/__init__.py", line 180, in <module>
    _site_initialize()
  File "/data/llvm_build_fffdd9ed/llvm/python_packages/mlir_core/mlir/_mlir_libs/__init__.py", line 78, in _site_initialize
    from ._mlir import ir
ImportError: cannot import name 'ir' from 'mlir._mlir_libs._mlir' (unknown location)

On the contrary, it works correctly if I use an LLVM build that was built on the local machine.
Is this expected? Do the two machines need to have exactly the same version of Python for this to work?

Thank you in advance