ARROW-8026: [Python] Support memoryview as a value type for creating …

…binary-like arrays

This also handles non-contiguous memory views.

Closes #6574 from wesm/ARROW-8026

Authored-by: Wes McKinney <wesm+git@apache.org>
Signed-off-by: Wes McKinney <wesm+git@apache.org>

cpp/src/arrow/python/common.h

@@ -257,6 +257,14 @@ struct PyBytesView {
       this->size = PyByteArray_GET_SIZE(obj);
       this->ref.reset();
       return Status::OK();
+    } else if (PyMemoryView_Check(obj)) {
+      PyObject* contig_view = PyMemoryView_GetContiguous(obj, PyBUF_READ, 'C');
+      RETURN_IF_PYERROR();
+      this->ref.reset(contig_view);
+      Py_buffer* buf = PyMemoryView_GET_BUFFER(contig_view);
+      this->bytes = reinterpret_cast<const char*>(buf->buf);
+      this->size = buf->len;
+      return Status::OK();
     } else {
       return Status::TypeError("Expected ", expected_msg, ", got a '",
                                Py_TYPE(obj)->tp_name, "' object");
@@ -266,10 +274,6 @@ struct PyBytesView {
   OwnedRef ref;
 };
 
-// Return the common PyArrow memory pool
-ARROW_PYTHON_EXPORT void set_default_memory_pool(MemoryPool* pool);
-ARROW_PYTHON_EXPORT MemoryPool* get_memory_pool();
-
 class ARROW_PYTHON_EXPORT PyBuffer : public Buffer {
  public:
   /// While memoryview objects support multi-dimensional buffers, PyBuffer only supports
@@ -285,6 +289,10 @@ class ARROW_PYTHON_EXPORT PyBuffer : public Buffer {
   Py_buffer py_buf_;
 };
 
+// Return the common PyArrow memory pool
+ARROW_PYTHON_EXPORT void set_default_memory_pool(MemoryPool* pool);
+ARROW_PYTHON_EXPORT MemoryPool* get_memory_pool();
+
 // This is annoying: because C++11 does not allow implicit conversion of string
 // literals to non-const char*, we need to go through some gymnastics to use
 // PyObject_CallMethod without a lot of pain (its arguments are non-const

cpp/src/arrow/python/helpers.h

@@ -78,7 +78,7 @@ ARROW_PYTHON_EXPORT
 bool PyFloat_IsNaN(PyObject* obj);
 
 inline bool IsPyBinary(PyObject* obj) {
-  return PyBytes_Check(obj) || PyByteArray_Check(obj);
+  return PyBytes_Check(obj) || PyByteArray_Check(obj) || PyMemoryView_Check(obj);
 }
 
 // \brief Convert a Python integer into a C integer

python/pyarrow/tests/test_convert_builtin.py

@@ -711,16 +711,19 @@ def test_large_binary_value(ty):
 
 def test_sequence_bytes():
     u1 = b'ma\xc3\xb1ana'
+
     data = [b'foo',
+            memoryview(b'dada'),
+            memoryview(b'd-a-t-a')[::2],  # non-contiguous is made contiguous
             u1.decode('utf-8'),  # unicode gets encoded,
             bytearray(b'bar'),
             None]
     for ty in [None, pa.binary(), pa.large_binary()]:
         arr = pa.array(data, type=ty)
-        assert len(arr) == 4
+        assert len(arr) == 6
         assert arr.null_count == 1
         assert arr.type == ty or pa.binary()
-        assert arr.to_pylist() == [b'foo', u1, b'bar', None]
+        assert arr.to_pylist() == [b'foo', b'dada', b'data', u1, b'bar', None]
 
 
 @pytest.mark.parametrize("ty", [pa.string(), pa.large_string()])