[C++] cumulative_sum ignores CumulativeSumOptions::check_overflow #35789
Comments
pitrou
pushed a commit
that referenced
this issue
Jun 1, 2023
### Rationale for this change
There are two variants of cumsum, `cumulative_sum` always wraps around on overflow and `cumulative_sum_checked` always return `Status::Invalid` on overflow, regardless of the `check_overflow` parameter in CumulativeSumOptions.
For example:
```cpp
CumulativeSumOptions options(/*start=*/0, /*skip_nulls=*/true, /*check_overflow=*/true);
auto res = CallFunction("cumulative_sum", {ArrayFromJSON(int8(), "[127, 1]")}, &options);
std::cout << res->make_array()->ToString() << std::endl;
// prints [127, -128], but user might expect an error returned.
```
I believe it's more of a ambiguity rather than a bug. The document of `cumulative_sum` also doesn't mention `check_overflow` at all, and `check_overflow` is not exposed to pyarrow. So IMO the best approach to avoid confusion is to remove `check_overflow` from CumulativeSumOptions.
The only place where `check_overflow` is used is in the C++ convenience function `CumulativeSum` defined in api_vector.h, which calls `cumulative_sum_checked` or `cumulative_sum` depending on the parameter. It can be replaced by a bool parameter in the C++ convenience function.
### What changes are included in this PR?
1. `check_overflow` is removed from CumulativeSumOptions.
2. Add a bool check_overflow parameter to the C++ convenience function CumulativeSum.
### Are these changes tested?
It doesn't affect the current tests.
### Are there any user-facing changes?
Yes, but only the C++ convenience function because check_overflow is not used in the kernel implementation and not exposed to pyarrow anyways.
* Closes: #35789
Authored-by: Jin Shang <[email protected]>
Signed-off-by: Antoine Pitrou <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Describe the bug, including details regarding any error messages, version, and platform.
There are two variants of cumsum,
cumulative_sumalways wraps around on overflow andcumulative_sumalways returnStatus::Invalidon overflow, regardless of thecheck_overflowparameter in CumulativeSumOptions.For example:
I believe it's more of a ambiguity rather than a bug. The document of
cumulative_sumalso doesn't memtioncheck_overflowat all, andcheck_overflowis not exposed to pyarrow. So IMO the best approach to avoid confusion is to removecheck_overflowfrom CumulativeSumOptions.The only place where
check_overflowis used is in the convenience functionCumulativeSumdefined in api_vector.h, which callscumulative_sum_checkedorcumulative_sumdepending on the parameter. It can be replaced by a bool parameter in the convenience function.Component(s)
C++
The text was updated successfully, but these errors were encountered: