Substrait integration doesn't recognize typed functions about arrow-datafusion HOT 7 OPEN

Cool, I'll take a look in the next days!

from arrow-datafusion.

Sounds great!! I'm very interested in this project, looking forward to your progress!

from arrow-datafusion.

Thanks for the report @Blizzara -- this would be a great thing to fix

from arrow-datafusion.

Thanks for taking this @Blizzara! Just out of curiosity, are you using this datafusion-substrait somewhere and find this inconsistency?

from arrow-datafusion.

Yes! I'm working on using DataFusion to basically execute Spark dataframes through Spark -> Substrait -> DataFusion. The Spark -> Substrait part is a (currently closed-source, but I hope to open it too) fork of "substrait-spark" from https://github.com/apache/incubator-gluten/tree/v1.1.0/substrait/substrait-spark (forked as it's no longer included in gluten).

from arrow-datafusion.

Be wary, Gluten contains a copy of Substrait instead of depending on the main repo. As a result its Substrait is incompatible with the rest of the ecosystem. That works for them as they only use the Substrait internally but the other tools are handy sometimes (especially the Substrait Validator).

It may also interest you that a generic Spark to Substrait tool is out there: https://github.com/voltrondata/spark-substrait-gateway

It's mostly there for DuckDB but other backends require some tweaks (DataFusion is one of them). We have a tweak to remove the compound names for DataFusion but would love to see this issue addressed. If you have questions (especially from the Substrait side) feel free to reach out.

from arrow-datafusion.

Be wary, Gluten contains a copy of Substrait instead of depending on the main repo. As a result its Substrait is incompatible with the rest of the ecosystem. That works for them as they only use the Substrait internally but the other tools are handy sometimes (especially the Substrait Validator).

Yup, the substrait-spark submodule was not using Gluten's Substrait but the vanilla one. (The whole thing wasn't really used by Gluten and it's no longer part of the repo, thus the fork).

It may also interest you that a generic Spark to Substrait tool is out there: https://github.com/voltrondata/spark-substrait-gateway

It does, thanks! Not directly useful as we need a solution in Java, but it's always good to have more references to look at.

from arrow-datafusion.