[Python] get_include() gives wrong directory in conda environment

[asfimport]

get_include seems to do:

 

def get_include():
    """
    Return absolute path to directory containing Arrow C++ include
    headers. Similar to numpy.get_include
    """
    return _os.path.join(_os.path.dirname(__file__), 'include') 

This returns something like:

/path/to/myconda/envs/envname/lib/python3.8/site-packages/pyarrow/include

which does not exist in a conda environment. The path where the headers actually get installed is to:

 

$ echo $CONDA_PREFIX
/path/to/myconda/envs/envname

$ ls $CONDA_PREFIX/include/arrow | head
adapters
api.h
array
array.h
buffer_builder.h
buffer.h
builder.h
c
chunked_array.h
chunk_resolver.h

Environment: conda
Reporter: Left Screen

Related issues:

• [Python] PyArrow C++ header files no longer always included in installed pyarrow (is related to)

_{Note: This issue was originally created as ARROW-18129. Please see the migration documentation for further details.}

[asfimport]

Alenka Frim / @AlenkaF:
Yes, in case of using conda the Arrow C++ headers are not found in the python package ({}pyarrow/include){} due to conda recipe using {}PYARROW_BUNDLE_ARROW_CPP_HEADERS=0{}, see:

arrow/dev/tasks/conda-recipes/arrow-cpp/build-pyarrow.sh

Line 11 in 16fd7f3

export PYARROW_BUNDLE_ARROW_CPP_HEADERS=0

With Apache Arrow 10.0.0 the C++ code connected to PyArrow (PyArrow C\) has been moved to the python package and so the header files for PyArrow C\ should always be included in the package itself. To track that work we created another issue https://issues.apache.org/jira/browse/ARROW-18340.

Regarding the headers for Arrow C\: we should add info in the documentation/docstrings to make it clear that the  Arrow C\ header files are not located in the python package if one is using conda.

We could also add $CONDA_PREFIX/include/arrow path in the get_include() function if one is using conda. @pitrou what is your opinion on that?

[jakirkham]

Is this still an issue?