Fix projected fields predicate evaluation

Author: Roman Shanin

pyiceberg/expressions/visitors.py

@@ -870,9 +870,10 @@ class _ColumnNameTranslator(BooleanExpressionVisitor[BooleanExpression]):
     file_schema: Schema
     case_sensitive: bool
 
-    def __init__(self, file_schema: Schema, case_sensitive: bool) -> None:
+    def __init__(self, file_schema: Schema, case_sensitive: bool, projected_missing_fields: dict[str, Any]) -> None:
         self.file_schema = file_schema
         self.case_sensitive = case_sensitive
+        self.projected_missing_fields = projected_missing_fields
 
     def visit_true(self) -> BooleanExpression:
         return AlwaysTrue()
@@ -894,12 +895,17 @@ def visit_unbound_predicate(self, predicate: UnboundPredicate[L]) -> BooleanExpr
 
     def visit_bound_predicate(self, predicate: BoundPredicate[L]) -> BooleanExpression:
         file_column_name = self.file_schema.find_column_name(predicate.term.ref().field.field_id)
+        field_name = predicate.term.ref().field.name
 
         if file_column_name is None:
             # In the case of schema evolution, the column might not be present
             # in the file schema when reading older data
             if isinstance(predicate, BoundIsNull):
                 return AlwaysTrue()
+            # Projected fields are only available for identity partition fields
+            # Which mean that partition pruning excluded partition field which evaluates to false
+            elif field_name in self.projected_missing_fields:
+                return AlwaysTrue()
             else:
                 return AlwaysFalse()
 
@@ -913,8 +919,8 @@ def visit_bound_predicate(self, predicate: BoundPredicate[L]) -> BooleanExpressi
             raise ValueError(f"Unsupported predicate: {predicate}")
 
 
-def translate_column_names(expr: BooleanExpression, file_schema: Schema, case_sensitive: bool) -> BooleanExpression:
-    return visit(expr, _ColumnNameTranslator(file_schema, case_sensitive))
+def translate_column_names(expr: BooleanExpression, file_schema: Schema, case_sensitive: bool, projected_missing_fields: dict[str, Any]) -> BooleanExpression:
+    return visit(expr, _ColumnNameTranslator(file_schema, case_sensitive, projected_missing_fields))
 
 
 class _ExpressionFieldIDs(BooleanExpressionVisitor[Set[int]]):

pyiceberg/io/pyarrow.py

@@ -1404,18 +1404,18 @@ def _task_to_record_batches(
         # the table format version.
         file_schema = pyarrow_to_schema(physical_schema, name_mapping, downcast_ns_timestamp_to_us=True)
 
-        pyarrow_filter = None
-        if bound_row_filter is not AlwaysTrue():
-            translated_row_filter = translate_column_names(bound_row_filter, file_schema, case_sensitive=case_sensitive)
-            bound_file_filter = bind(file_schema, translated_row_filter, case_sensitive=case_sensitive)
-            pyarrow_filter = expression_to_pyarrow(bound_file_filter)
-
         # Apply column projection rules
         # https://iceberg.apache.org/spec/#column-projection
         should_project_columns, projected_missing_fields = _get_column_projection_values(
             task.file, projected_schema, partition_spec, file_schema.field_ids
         )
 
+        pyarrow_filter = None
+        if bound_row_filter is not AlwaysTrue():
+            translated_row_filter = translate_column_names(bound_row_filter, file_schema, case_sensitive=case_sensitive, projected_missing_fields=projected_missing_fields)
+            bound_file_filter = bind(file_schema, translated_row_filter, case_sensitive=case_sensitive)
+            pyarrow_filter = expression_to_pyarrow(bound_file_filter)
+
         file_project_schema = prune_columns(file_schema, projected_field_ids, select_full_types=False)
 
         fragment_scanner = ds.Scanner.from_fragment(