fix(optimizer)!: query schema directly when type annotation fails for processing UNNEST source #6451
Conversation
|
can you fix the type annotation instead? |
|
Yeah, I second that suggestion in favor of avoiding making this more complicated than necessary. Keen to understand more if that is tricky to do; let me know and we can hop in a call and chat about this. |
georgesittas
left a comment
georgesittas
left a comment
There was a problem hiding this comment.
@fivetran-BradfordPaskewitz keep in mind that you'll need to rebase and apply these changes to resolver.py due to 625654a.
fivetran-BradfordPaskewitz
force-pushed
the
fix_qualify_unnest
branch
2 times, most recently
from
3b0a522 to
fea0204
Compare
fivetran-BradfordPaskewitz
force-pushed
the
fix_qualify_unnest
branch
from
fea0204 to
41cfa9e
Compare
georgesittas
left a comment
georgesittas
left a comment
There was a problem hiding this comment.
@fivetran-BradfordPaskewitz another quick round of comments from me, but this should be good to merge soon.
@tobymao wanna take a look as well? Another pair of eyes is a good idea for this one.
sqlglot/optimizer/resolver.py
Outdated
| if col_type and col_type.is_type(exp.DataType.Type.ARRAY): | ||
| element_types = col_type.expressions | ||
| if element_types: | ||
| unnest.type = element_types[0] |
There was a problem hiding this comment.
I think we should copy here?
There was a problem hiding this comment.
done
sqlglot/optimizer/resolver.py
Outdated
| if element_types: | ||
| unnest.type = element_types[0] | ||
| else: | ||
| unnest.type = col_type |
There was a problem hiding this comment.
done
sqlglot/optimizer/resolver.py
Outdated
| return None | ||
| table_name = table_identifier.name | ||
|
|
||
| source = scope.sources[table_name] |
There was a problem hiding this comment.
Is this lookup safe? Do we always know that table_name appears in sources? Should we be conservative?
There was a problem hiding this comment.
changed this to be defensive
sqlglot/optimizer/resolver.py
Outdated
| return self._get_column_type_from_scope(source, column.name) | ||
|
|
||
| def _get_column_type_from_scope( | ||
| self, source: t.Union[Scope, exp.Table], col_name: str |
There was a problem hiding this comment.
We should pass a Column instance here, not col_name, because columns will already be normalized & quoted and we won't risk weird normalization corner cases when looking them up in the schema– which also normalizes.
There was a problem hiding this comment.
good call, updated to use column instead of name
sqlglot/optimizer/resolver.py
Outdated
| for source_name in source.sources: | ||
| nested_source = source.sources[source_name] |
There was a problem hiding this comment.
| for source_name in source.sources: | |
| nested_source = source.sources[source_name] | |
| for source_name, nested_source in source.sources.items(): |
There was a problem hiding this comment.
updated
… processing UNNEST source
fivetran-BradfordPaskewitz
force-pushed
the
fix_qualify_unnest
branch
from
41cfa9e to
7428a39
Compare
Issue: https://fivetran.atlassian.net/browse/RD-1066808
Problem: In BigQuery, when you UNNEST an ARRAY<STRUCT<...>>, the struct fields can be referenced as unqualified columns (e.g., WHERE type = 'x'). This worked in simple queries but failed in correlated subqueries because sqlglot's type annotation couldn't resolve columns from outer scopes, so qualify_columns didn't know what struct fields to expose.
Solution: Add a fallback mechanism in qualify_columns.py that queries the schema directly when type annotation fails. When processing an UNNEST source, if the expression isn't typed, traverse parent scopes to find the column's type definition in the schema (handling both table sources and CTEs), then extract and expose the struct field names as available columns in that scope.