4 ms·

Sparrow: C++20 Idiomatic APIs for the Apache Arrow Columnar Format

36 points by tanelpoder 1y ago

mgaunard 1y ago
The official arrow implementation is already in C++11, not sure what the value proposition of this is.
- jcelerier 1y ago
  from a quick glance, https://man-group.github.io/sparrow/basic_usage.html https://man-group.github.io/sparrow/basic_usage.html looks infinitely more like a library I want to use than https://arrow.apache.org/cookbook/cpp/basic.html https://arrow.apache.org/cookbook/cpp/basic.html
- rfoo 1y ago
  <rant>The official Arrow C++ implementation is just ergonomic warts, full of `const std::shared_ptr<T>&` bs. Trying to use it to manipulate data always give me headache telling apart WTH is an Array, ArrayData, Buffer, and the typed Array interfaces are barely usable. The original official Rust port inherited all the mis-designs too. On the Rust side someone created arrow2 [0] to fix it.</rant> And I'm glad there's a good C++ impl too. [0] https://github.com/jorgecarleitao/arrow2 https://github.com/jorgecarleitao/arrow2
  - mgaunard 1y ago
    that's because a given Arrow column is actually several arrays of arrays. Array, ArrayData and Buffer map to different layers of the abstraction.
levzettelin 1y ago
// You are responsible for releasing the structure in the end arrow_array.release(&arrow_array); This doesn't look like RAII. How is this idiomatic for C++20? Why do you have to pass a pointer to "this" again as an explicit argument.
- rfoo 1y ago
  This is the extracted Arrow C data interfaces as documented in https://arrow.apache.org/docs/format/CDataInterface.html https://arrow.apache.org/docs/format/CDataInterface.html It's not how you interact with the data in your own C++ code, it's for passing this data to other in-process consumers (libraries etc). While in the example it calls the release function, this is usually just passed to a downstream consumer and it's their responsibility to call it. I agree that having such an example as the first one is confusing. Given that a large part of the point of Apache Arrow is passing data columnar data between libraries in different languages in memory, it makes some sense.
  - CyberDildonics 1y ago
    It's not how you interact with the data in your own C++ code, it's for passing this data to other in-process consumers (libraries etc). While in the example it calls the release function, this is usually just passed to a downstream consumer and it's their responsibility to call it. This seems like a strange rationalizations when you don't need to have explicit release to be able to pass it to something else.
- pjmlp 1y ago
  RAII predates C++98, I was already used to it in Turbo C++ for MS-DOS, and is pity we need to keep advocating for it as something extraordinary.
  - CyberDildonics 1y ago
    This doesn't have anything to do with what they said, they didn't say RAII was new.
    - pjmlp 1y ago
      Might be misunderstood by others not skilled in C++ when reading, > This doesn't look like RAII. How is this idiomatic for C++20?

36 points by tanelpoder 1y ago

mgaunard 1y ago

The official arrow implementation is already in C++11, not sure what the value proposition of this is.

jcelerier 1y ago
from a quick glance, https://man-group.github.io/sparrow/basic_usage.html https://man-group.github.io/sparrow/basic_usage.html looks infinitely more like a library I want to use than https://arrow.apache.org/cookbook/cpp/basic.html https://arrow.apache.org/cookbook/cpp/basic.html
rfoo 1y ago
<rant>The official Arrow C++ implementation is just ergonomic warts, full of `const std::shared_ptr<T>&` bs. Trying to use it to manipulate data always give me headache telling apart WTH is an Array, ArrayData, Buffer, and the typed Array interfaces are barely usable. The original official Rust port inherited all the mis-designs too. On the Rust side someone created arrow2 [0] to fix it.</rant> And I'm glad there's a good C++ impl too. [0] https://github.com/jorgecarleitao/arrow2 https://github.com/jorgecarleitao/arrow2
- mgaunard 1y ago
  that's because a given Arrow column is actually several arrays of arrays. Array, ArrayData and Buffer map to different layers of the abstraction.

levzettelin 1y ago

// You are responsible for releasing the structure in the end arrow_array.release(&arrow_array); This doesn't look like RAII. How is this idiomatic for C++20? Why do you have to pass a pointer to "this" again as an explicit argument.

rfoo 1y ago
This is the extracted Arrow C data interfaces as documented in https://arrow.apache.org/docs/format/CDataInterface.html https://arrow.apache.org/docs/format/CDataInterface.html It's not how you interact with the data in your own C++ code, it's for passing this data to other in-process consumers (libraries etc). While in the example it calls the release function, this is usually just passed to a downstream consumer and it's their responsibility to call it. I agree that having such an example as the first one is confusing. Given that a large part of the point of Apache Arrow is passing data columnar data between libraries in different languages in memory, it makes some sense.
- CyberDildonics 1y ago
  It's not how you interact with the data in your own C++ code, it's for passing this data to other in-process consumers (libraries etc). While in the example it calls the release function, this is usually just passed to a downstream consumer and it's their responsibility to call it. This seems like a strange rationalizations when you don't need to have explicit release to be able to pass it to something else.
pjmlp 1y ago
RAII predates C++98, I was already used to it in Turbo C++ for MS-DOS, and is pity we need to keep advocating for it as something extraordinary.
- CyberDildonics 1y ago
  This doesn't have anything to do with what they said, they didn't say RAII was new.
  - pjmlp 1y ago
    Might be misunderstood by others not skilled in C++ when reading, > This doesn't look like RAII. How is this idiomatic for C++20?