Hi all, I'm interested in making it possible to convert NumPy variable-width string data to arrow strings. See https://github.com/apache/arrow/issues/42018. In particular, I want to fix this:
>>> import numpy as np >>> import pyarrow as pa >>> arr = np.array(["hello", "world"], dtype="T") >>> pa.array(arr) pyarrow.lib.ArrowNotImplementedError: Unsupported numpy type 2056 There was an effort to get this fixed last year ( https://github.com/apache/arrow/pull/48391) but it stalled principally because supporting NumPy's StringDType in Arrow requires targeting the NumPy 2.0 C API, see https://github.com/apache/arrow/issues/48473. There is an open PR that updates NumPy 2.0 to the minimum supported NumPy version, https://github.com/apache/arrow/pull/50769. I attended the PyArrow dev meeting today and Joris mentioned that we should ping the mailing list about this. If anyone sees a problem with updating the minimum supported NumPy version, we'd appreciate it if you could comment here or on the issue. It may be possible to keep support for older NumPy versions working, at the cost of some internal complexity so it would help to hear about real-world usecases to see if the internal complexity is worth it. -Nathan
