Rewrite conversion in terms of column #15213
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
It looks like soon after I started investigating scalar conversions for #14121 (but well before I made the PR) a major underlying hole was plugged in pyarrow via apache/arrow#36162. Most of #14121 was created to give us a way to handle scalars from pyarrow generically in libcudf. Now that pyarrow scalars can be easily tossed into arrays, we no longer really need separate scalar functions in libcudf; we can simply create an array from the scalar, put it into a table, and then call the table function.
Additionally, arrow also has a function for creating an array from a scalar. This function is not new but was previously undocumented. The builder code added to libcudf in #14121 can be removed and replaced with that factory. The scalar conversion is as simple as calling that arrow function and then using our preexisting
from_arrow
function on the resulting array.For now this PR is just a simplification of internals. Future PRs will remove the scalar API once we have a more standard path for the conversion of arrays via the C Data Interface.
Checklist