Experiment with Extension as a DataType #7398
Draft
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
parquet
paleolimbot
changed the title
tustvold
reviewed
Apr 9, 2025
| @@ -276,6 +276,7 @@ fn build_extend(array: &ArrayData) -> Extend { | |||
| UnionMode::Dense => union::build_extend_dense(array), | |||
| }, | |||
| DataType::RunEndEncoded(_, _) => todo!(), | |||
| DataType::Extension(_) => unimplemented!("Extension not implemented"), | |||
There was a problem hiding this comment.
This highlights the challenge with this approach, every kernel/function must now be updated to accommodate this new extension type. Not only does this stand a very high chance of regressing functionality, but also is IMO not in the spirit of extension types.
There was a problem hiding this comment.
Yes, the tradeoff is taking on the burden of extensibility (so that other projects don't have to implement it separately, for example, by redefining wrapper DataTypes or Arrays). This extra burden hasn't caused issues in Arrow C++/pyarrow that I'm aware of but absolutely fair that it might here.
This was referenced Apr 9, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Which issue does this PR close?
Experiments in pursuit of #7063 and apache/datafusion#12644 .
Rationale for this change
Customizing behaviour (e.g., sorting, signatures for user defined functions, import/export from/to formats like Parquet, Arrow IPC, and FFI) is increasingly requested. arrow-rs is also increasingly the implementation of choice for compute frameworks and many APIs have been built around the
DataTypeandArrayRef(e.g., GeoArrow, parquet, many parts of DataFusion). To support additional types, those frameworks have to either invent a new array type and rewrite substantial wrappers (e.g., GeoArrow), or use something other than aDataType(e.g., a Field or MyOwnDataType) and forego having that type automatically pass through APIs from other arrow-rs-based libraries (or arrow itself).Incorporating an ExtensionType as a first-class data type provides a potentially less disruptive route for libraries like parquet and DataFusion to implement new Parquet types and other types other databases implement like JSON and UUID. I also have an experiment going for swapping out a
Fieldfor aDataTypein DataFusion ( apache/datafusion#15036 ).What changes are included in this PR?
A new
DataType::Extension()enum member was added. I had hoped to have this beDataType::Extension(Arc<dyn ExtensionType>)but the current design of the ExtensionType is not dyn-compatible. TheDynExtensionTypetrait is really a placeholder to collect the requirements of the rest of arrow-rs for this experiment.This is currently just an experiment/minimum change that actually builds! I'm happy to continue the experiment if there is interest.
Are there any user-facing changes?
Yes, a new data type enum member at least (which is a breaking change).