-
Notifications
You must be signed in to change notification settings - Fork 4
Open
Labels
documentationImprovements or additions to documentationImprovements or additions to documentationeffort - highmajor issue that will require multiple steps or complex designmajor issue that will require multiple steps or complex designoptimizationImproving the speed/quality of PyDough's outputsImproving the speed/quality of PyDough's outputs
Description
We need to maintain a forked version of SQLGlot to address incompatibilities between our requirements and the standard SQLGlot optimizer.
We're currently skipping problematic rules:
- The
canonicalizerule (12) removes nested DATE functions, causing data loss:DATE(DATE(a,b),c) -> DATE(a,b) # 'c' argument disappears merge_subqueriesrules break ANTI/SEMI joins if not conditionally skipped.unnest_subqueriescreates tries to remove ANTI/SEMI join but creates aliases with no corresponding tables.
Proposed Solution
Fork SQLGlot to fix these optimizer bugs directly in the source, allowing us to use the optimizer without skipping critical rules.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
documentationImprovements or additions to documentationImprovements or additions to documentationeffort - highmajor issue that will require multiple steps or complex designmajor issue that will require multiple steps or complex designoptimizationImproving the speed/quality of PyDough's outputsImproving the speed/quality of PyDough's outputs