@@ -71,7 +71,7 @@ use crate::parquet::schema_adapter::{
7171use arrow:: datatypes:: { DataType , FieldRef , Schema , SchemaRef } ;
7272use async_trait:: async_trait;
7373use bytes:: Bytes ;
74- use datafusion:: common:: Result as DFResult ;
74+ use datafusion:: common:: { DataFusionError , Result as DFResult } ;
7575use datafusion:: datasource:: physical_plan:: parquet:: metadata:: DFParquetMetadata ;
7676use datafusion:: datasource:: physical_plan:: parquet:: {
7777 ParquetFileMetrics , ParquetFileReaderFactory ,
@@ -668,7 +668,12 @@ impl AsyncFileReader for EagerPageIndexReader {
668668 Arc :: new ( schema) ,
669669 conversion_options. clone ( ) ,
670670 )
671- . map_err ( |e| ParquetError :: External ( Box :: new ( e) ) ) ?;
671+ . map_err ( |e| match e {
672+ // Keep structured Spark errors directly inside the Parquet wrapper so
673+ // the JNI error converter preserves Spark's exception type and class.
674+ DataFusionError :: External ( source) => ParquetError :: External ( source) ,
675+ other => ParquetError :: External ( Box :: new ( other) ) ,
676+ } ) ?;
672677 if * row_filters {
673678 if let Some ( rejection) = & rejection {
674679 self . deferred_rejections . lock ( ) . unwrap ( ) . insert (
@@ -1308,6 +1313,64 @@ mod tests {
13081313 assert ! ( metadata_for( false , without_ids) . await . is_ok( ) ) ;
13091314 }
13101315
1316+ #[ tokio:: test]
1317+ async fn conversion_check_preserves_structured_spark_errors ( ) {
1318+ let store = Arc :: new ( InMemory :: new ( ) ) ;
1319+ let schema = Arc :: new ( Schema :: new ( vec ! [
1320+ arrow:: datatypes:: Field :: new( "B" , DataType :: Int32 , false ) ,
1321+ arrow:: datatypes:: Field :: new( "b" , DataType :: Int32 , false ) ,
1322+ ] ) ) ;
1323+ let values: arrow:: array:: ArrayRef = Arc :: new ( Int32Array :: from ( vec ! [ 1 , 2 ] ) ) ;
1324+ let batch =
1325+ RecordBatch :: try_new ( Arc :: clone ( & schema) , vec ! [ Arc :: clone( & values) , values] ) . unwrap ( ) ;
1326+ let mut bytes = Vec :: new ( ) ;
1327+ let mut writer = ArrowWriter :: try_new ( & mut bytes, schema, None ) . unwrap ( ) ;
1328+ writer. write ( & batch) . unwrap ( ) ;
1329+ writer. close ( ) . unwrap ( ) ;
1330+ let size = bytes. len ( ) as u64 ;
1331+ let location = "duplicates.parquet" ;
1332+ store
1333+ . put ( & Path :: from ( location) , Bytes :: from ( bytes) . into ( ) )
1334+ . await
1335+ . unwrap ( ) ;
1336+ let runtime = datafusion:: execution:: runtime_env:: RuntimeEnv :: default ( ) ;
1337+ let metrics = ExecutionPlanMetricsSet :: new ( ) ;
1338+ let required = Arc :: new ( Schema :: new ( vec ! [ arrow:: datatypes:: Field :: new(
1339+ "B" ,
1340+ DataType :: Int32 ,
1341+ false ,
1342+ ) ] ) ) ;
1343+ let mut options =
1344+ SparkParquetOptions :: new ( datafusion_comet_spark_expr:: EvalMode :: Legacy , "UTC" , false ) ;
1345+ options. case_sensitive = false ;
1346+ let factory = EagerPageIndexReaderFactory :: new (
1347+ store,
1348+ runtime. cache_manager . get_file_metadata_cache ( ) ,
1349+ ScanIoSource :: Local ,
1350+ & metrics,
1351+ )
1352+ . with_conversion_check ( required, options, true ) ;
1353+ let mut reader = factory
1354+ . create_reader (
1355+ 0 ,
1356+ PartitionedFile :: new ( location. to_string ( ) , size) ,
1357+ None ,
1358+ & metrics,
1359+ )
1360+ . unwrap ( ) ;
1361+ let error = reader. get_metadata ( None ) . await . unwrap_err ( ) ;
1362+ let ParquetError :: External ( source) = error else {
1363+ panic ! ( "expected structured Spark error, got {error}" ) ;
1364+ } ;
1365+ assert ! (
1366+ matches!(
1367+ source. downcast_ref:: <SparkError >( ) ,
1368+ Some ( SparkError :: DuplicateFieldCaseInsensitive { .. } )
1369+ ) ,
1370+ "unexpected error: {source}"
1371+ ) ;
1372+ }
1373+
13111374 #[ test]
13121375 fn variant_policy_preserves_footer_metadata_and_indexes ( ) {
13131376 let schema = Arc :: new ( Schema :: new_with_metadata (
0 commit comments