llama90 commented on issue #35560:
URL: https://github.com/apache/arrow/issues/35560#issuecomment-1837151416
Hello,
I am interested in this issue and looking into it.
First, I found functions that are using the legacy `Scalar.CastTo()` in the
code.
I have reorganized these functions to call the `Cast` compute kernel where
tests are possible.
| file name | code snippet
| note |
|-------------------|--------------------------------------------------------------------------------------------|------|
| file_parquet.cc | `auto maybe_min = min->CastTo(field.type());`
| - |
| file_parquet.cc | `auto maybe_max = max->CastTo(field.type());`
| - |
| partition_test.cc | `ASSERT_OK_AND_ASSIGN(auto dict_hello,
MakeScalar("hello")->CastTo(DictStr("")->type()));` | 1 |
| partition_test.cc | `ASSERT_OK_AND_ASSIGN(auto dict_hello,
MakeScalar("hello")->CastTo(dict_type));` | 1 |
| partition_test.cc | `ASSERT_OK_AND_ASSIGN(auto day,
StringScalar("2020-06-08").CastTo(temporal));` | ✅ |
| scalar.cc | `auto maybe_repr = CastTo(utf8());`
| - |
| scalar.cc | `ARROW_ASSIGN_OR_RAISE(auto cast_value,
from_.CastTo(dict_type.value_type()));` | - |
| scalar.cc | `return
Int32Scalar(0).CastTo(dict_type.index_type()).Value(&out.index);`
| - |
| scalar.h | `Result<std::shared_ptr<Scalar>>
CastTo(std::shared_ptr<DataType> to) const;` | - |
| scalar_test.cc | `TimestampScalar(value,
timestamp(in)).CastTo(timestamp(out)).ValueOrDie();` | 2 |
| scalar_test.cc | `TimestampScalar(1024,
timestamp(TimeUnit::MILLI)).CastTo(utf8()));` | 2 |
| scalar_test.cc | `TimestampScalar(1024,
timestamp(TimeUnit::MILLI)).CastTo(int64()));` | 2 |
| scalar_test.cc | `.CastTo(date64()));`
| 2 |
| scalar_test.cc | `ASSERT_OK_AND_ASSIGN(auto cast_to_other,
scalar->CastTo(other_type))` | ✅ |
| scalar_test.cc | `ASSERT_OK_AND_ASSIGN(auto cast_from_other,
other_scalar->CastTo(type))` | ✅ |
| scalar_test.cc | `StringScalar(std::string(repr)).CastTo(type));`
| ✅ |
| scalar_test.cc | `ASSERT_OK_AND_ASSIGN(auto cast_to_string,
scalar->CastTo(utf8()));` | ✅ |
| scalar_test.cc | `EXPECT_OK_AND_ASSIGN(auto cast_scalar,
scalar.CastTo(to_type));` | 3 |
| scalar_test.cc | `scalar.CastTo(to_type));`
| 3 |
| scalar_test.cc | `ASSERT_OK_AND_ASSIGN(auto cast_alpha,
alpha->CastTo(ty));` | 1 |
* 1: Unsupported cast from string to dictionary using function
cast_dictionary
* 2: ValueOrDie called on an error: Invalid: Casting from timestamp[ns] to
timestamp[us] would lose data: 1234
* 3: Unsupported cast from list_view<item: int16> to list using function
cast_list
* **In some functions, the same test cases may yield different errors
depending on the additional cast kernel implementation**.
* This means that in some cases, the errors are indicated based on the
error messages triggered by the functions executed first.
Looking at the table, the calling functions can be organized as follows:
1. Cases where the cast operation is implemented and calling the Cast
function works without additional implementation.
2. Cases where additional implementation of the cast operation is needed.
For now, I have submitted a PR for the cases that are working happy case,
and I am curious if this aligns with the intentions mentioned in the issue.
If PR aligns with the intent, I would like to proceed by separating the
happy cases from those that require additional cast implementation, and I would
appreciate your feedback.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
