Neal Richardson created ARROW-9380:
--------------------------------------
Summary: [C++] Segfaults in compute::CallFunction
Key: ARROW-9380
URL: https://issues.apache.org/jira/browse/ARROW-9380
Project: Apache Arrow
Issue Type: Bug
Components: C++
Reporter: Neal Richardson
I triggered these from R, so that's what the reproducers are in.
1. Calling "filter" with no args segfaults.
{code:r}
arrow:::compute__CallFunction("filter", list(), list(keep_na = FALSE))
{code}
Top of the backtrace from lldb:
{code}
* frame #0: 0x0000000109e1c2c7 libarrow.100.dylib`arrow::Datum::type() const
+ 7
frame #1: 0x000000010a14a232
libarrow.100.dylib`arrow::compute::internal::(anonymous
namespace)::FilterMetaFunction::ExecuteImpl(std::__1::vector<arrow::Datum,
std::__1::allocator<arrow::Datum> > const&, arrow::compute::FunctionOptions
const*, arrow::compute::ExecContext*) const + 66
frame #2: 0x0000000109fc32c9
libarrow.100.dylib`arrow::compute::MetaFunction::Execute(std::__1::vector<arrow::Datum,
std::__1::allocator<arrow::Datum> > const&, arrow::compute::FunctionOptions
const*, arrow::compute::ExecContext*) const + 41
frame #3: 0x0000000109fb3d3c
libarrow.100.dylib`arrow::compute::CallFunction(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&,
std::__1::vector<arrow::Datum, std::__1::allocator<arrow::Datum> > const&,
arrow::compute::FunctionOptions const*, arrow::compute::ExecContext*) + 844
frame #4: 0x0000000109fb3c47
libarrow.100.dylib`arrow::compute::CallFunction(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&,
std::__1::vector<arrow::Datum, std::__1::allocator<arrow::Datum> > const&,
arrow::compute::FunctionOptions const*, arrow::compute::ExecContext*) + 599
{code}
This is not the case with at least some other functions. If I try to call "sum"
with no args, I get {{Invalid: Function accepts 1 arguments but passed 0}} and
no segfault.
2. Something is strange with is_null. It creates what appears to be a valid
boolean array, but if I pass it to filter, it segfaults. I'm adding bindings
for this in ARROW-9187 but this should run on current master:
{code:r}
library(arrow)
a <- Array$create(1:4)
b <- arrow:::shared_ptr(Array, arrow:::call_function("is_null", a))
a$Filter(b)
{code}
Backtrace:
{code}
* frame #0: 0x000000010a120bb6
libarrow.100.dylib`arrow::compute::internal::GetFilterOutputSize(arrow::ArrayData
const&, arrow::compute::FilterOptions::NullSelectionBehavior) + 38
frame #1: 0x000000010a125659
libarrow.100.dylib`arrow::compute::internal::(anonymous
namespace)::PrimitiveFilter(arrow::compute::KernelContext*,
arrow::compute::ExecBatch const&, arrow::Datum*) + 121
frame #2: 0x0000000109fbbea4
libarrow.100.dylib`arrow::compute::detail::VectorExecutor::ExecuteBatch(arrow::compute::ExecBatch
const&, arrow::compute::detail::ExecListener*) + 996
frame #3: 0x0000000109fba3e6
libarrow.100.dylib`arrow::compute::detail::VectorExecutor::Execute(std::__1::vector<arrow::Datum,
std::__1::allocator<arrow::Datum> > const&,
arrow::compute::detail::ExecListener*) + 150
frame #4: 0x0000000109fc0948
libarrow.100.dylib`arrow::compute::Function::Execute(std::__1::vector<arrow::Datum,
std::__1::allocator<arrow::Datum> > const&, arrow::compute::FunctionOptions
const*, arrow::compute::ExecContext*) const + 1016
frame #5: 0x0000000109fb3d3c
libarrow.100.dylib`arrow::compute::CallFunction(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&,
std::__1::vector<arrow::Datum, std::__1::allocator<arrow::Datum> > const&,
arrow::compute::FunctionOptions const*, arrow::compute::ExecContext*) + 844
frame #6: 0x000000010a14a9b5
libarrow.100.dylib`arrow::compute::internal::(anonymous
namespace)::FilterMetaFunction::ExecuteImpl(std::__1::vector<arrow::Datum,
std::__1::allocator<arrow::Datum> > const&, arrow::compute::FunctionOptions
const*, arrow::compute::ExecContext*) const + 1989
frame #7: 0x0000000109fc32c9
libarrow.100.dylib`arrow::compute::MetaFunction::Execute(std::__1::vector<arrow::Datum,
std::__1::allocator<arrow::Datum> > const&, arrow::compute::FunctionOptions
const*, arrow::compute::ExecContext*) const + 41
frame #8: 0x0000000109fb3d3c
libarrow.100.dylib`arrow::compute::CallFunction(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&,
std::__1::vector<arrow::Datum, std::__1::allocator<arrow::Datum> > const&,
arrow::compute::FunctionOptions const*, arrow::compute::ExecContext*) + 844
frame #9: 0x0000000109fb3c47
libarrow.100.dylib`arrow::compute::CallFunction(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&,
std::__1::vector<arrow::Datum, std::__1::allocator<arrow::Datum> > const&,
arrow::compute::FunctionOptions const*, arrow::compute::ExecContext*) + 599
{code}
BUT: if I call {{as.vector}} on {{b}} before using it as a Filter, it
works--even though I've discarded the as.vector result and am still using the
Array to filter.
{code:r}
library(arrow)
a <- Array$create(1:4)
b <- arrow:::shared_ptr(Array, arrow:::call_function("is_null", a))
as.vector(b)
a$Filter(b)
{code}
Just printing (calling {{ToString}}) on {{b}} doesn't prevent the segfault. And
I have not observed this with other boolean kernels. E.g. this does not
segfault:
{code:r}
library(arrow)
a <- Array$create(1:4)
b <- arrow:::shared_ptr(Array, arrow:::call_function("greater", a,
Scalar$create(3L)))
a$Filter(b)
{code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
