fix: migrate to new RDataSource column reader API - #4048
Conversation
Change the implementation of the generated RDataSource class to use the newer signature that returns one column reader. This, together with the changes at root-project/root#22234, allow to restore the behaviour that awkward relied upon to store arrays to disk via RDataFrame. Irrespective of these changes, the current in-memory layout of the awkward RecordArray is not fit for storage, leading to writing silently garbage data to disk. Fixes scikit-hep#3885
ianna
left a comment
There was a problem hiding this comment.
@vepadulano - Thank you so much for providing the fix! As we have discussed at the ROOT hackathon this should fix the Snapshot issue and migrate our implementation to the latest ROOT version. So we are (almost) at the state where we were before. I will schedule the PR for merge for the upcoming release. Thanks!
|
Removed specific version for root in create-args.
|
@ikrommyd - I’m currently actively coding and pushing changes to this branch to help move it forward. Could you hold off on merging |
Change the implementation of the generated RDataSource class to use the newer signature that returns one column reader. This, together with the changes at root-project/root#22234, allow to restore the behaviour that awkward relied upon to store arrays to disk via RDataFrame. Irrespective of these changes, the current in-memory layout of the awkward RecordArray is not fit for storage, leading to writing silently garbage data to disk.
Fixes #3885