Siyuan Zhuang created ARROW-3765:
------------------------------------
Summary: Gandiva segfault when using int64 recordbatch as its input
Key: ARROW-3765
URL: https://issues.apache.org/jira/browse/ARROW-3765
Project: Apache Arrow
Issue Type: Bug
Components: C++, Gandiva
Reporter: Siyuan Zhuang
This is because the `validity buffer` could be `None`:
{code}
>>> df = pd.DataFrame(np.random.randint(0, 100, size=(2**12, 10)))
>>> pa.Table.from_pandas(df).to_batches()[0].column(0).buffers()
[None, <pyarrow.lib.Buffer object at 0x110c1a228>]
>>> df = pd.DataFrame(np.random.randint(0, 100, size=(2**12, 10))*1.0)
>>> pa.Table.from_pandas(df).to_batches()[0].column(0).buffers()
[<pyarrow.lib.Buffer object at 0x11a2b3030>, <pyarrow.lib.Buffer object at
0x11a2b3228>]{code}
But Gandiva has not implemented it yet, thus accessing a nullptr:
{code}
void Annotator::PrepareBuffersForField(const FieldDescriptor& desc, const
arrow::ArrayData& array_data, EvalBatch* eval_batch) {
int buffer_idx = 0;
// TODO:
// - validity is optional
uint8_t* validity_buf =
const_cast<uint8_t*>(array_data.buffers[buffer_idx]->data());
eval_batch->SetBuffer(desc.validity_idx(), validity_buf);
++buffer_idx;
{code}
Reproduce code:
{code:java}
frame_data = np.random.randint(0, 100, size=(2**22, 10))
table = pa.Table.from_pandas(df)
filt = ... # Create any gandiva filter
r = filt.evaluate(table.to_batches()[0], pa.default_memory_pool()) #
segfault{code}
Backtrace:
{code:java}
* thread #2, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS
(code=1, address=0x10)
* frame #0: 0x00000001060184fc
libarrow.12.dylib`arrow::Buffer::data(this=0x0000000000000000) const at
buffer.h:162
frame #1: 0x0000000106fbed78
libgandiva.12.dylib`gandiva::Annotator::PrepareBuffersForField(this=0x0000000100624dc8,
desc=0x000000010101e138, array_data=0x000000010061f8e8,
eval_batch=0x0000000100796848) at annotator.cc:65
frame #2: 0x0000000106fbf4ed
libgandiva.12.dylib`gandiva::Annotator::PrepareEvalBatch(this=0x0000000100624dc8,
record_batch=0x00000001007a45b8, out_vector=size=1) at annotator.cc:94
frame #3: 0x00000001071449b7
libgandiva.12.dylib`gandiva::LLVMGenerator::Execute(this=0x0000000100624da0,
record_batch=0x00000001007a45b8, output_vector=size=1) at llvm_generator.cc:102
frame #4: 0x0000000107059a4f
libgandiva.12.dylib`gandiva::Filter::Evaluate(this=0x000000010079c668,
batch=0x00000001007a45b8,
out_selection=std::__1::shared_ptr<gandiva::SelectionVector>::element_type @
0x00000001007a43e8 strong=2 weak=1) at filter.cc:106
frame #5: 0x000000010948e002
gandiva.cpython-36m-darwin.so`__pyx_pw_7pyarrow_7gandiva_6Filter_3evaluate(_object*,
_object*, _object*) + 1986
frame #6: 0x0000000100140e8b Python`_PyCFunction_FastCallDict + 475
frame #7: 0x00000001001d28ca Python`call_function + 602
frame #8: 0x00000001001cf798 Python`_PyEval_EvalFrameDefault + 24616
frame #9: 0x00000001001d3cf9 Python`fast_function + 569
frame #10: 0x00000001001d2899 Python`call_function + 553
frame #11: 0x00000001001cf798 Python`_PyEval_EvalFrameDefault + 24616
frame #12: 0x00000001001d34c6 Python`_PyEval_EvalCodeWithName + 2902
frame #13: 0x00000001001c96e0 Python`PyEval_EvalCode + 48
frame #14: 0x00000001002029ae Python`PyRun_FileExFlags + 174
frame #15: 0x0000000100201f75 Python`PyRun_SimpleFileExFlags + 277
frame #16: 0x000000010021ef46 Python`Py_Main + 3558
frame #17: 0x0000000100000e08 Python`___lldb_unnamed_symbol1$$Python + 248
frame #18: 0x00007fff6ea72085 libdyld.dylib`start + 1{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
