Support invalid partToExtract for parse_url (#11661)

Signed-off-by: Haoyang Li <[email protected]>

integration_tests/src/main/python/url_test.py

@@ -148,7 +148,7 @@
 
 url_gen = StringGen(url_pattern)
 
-supported_parts = ['PROTOCOL', 'HOST', 'QUERY', 'PATH']
+supported_parts = ['PROTOCOL', 'HOST', 'QUERY', 'PATH', 'invalid', 'path']
 unsupported_parts = ['REF', 'FILE', 'AUTHORITY', 'USERINFO']
 
 @pytest.mark.parametrize('data_gen', [url_gen, edge_cases_gen], ids=idfn)

sql-plugin/src/main/scala/com/nvidia/spark/rapids/GpuOverrides.scala

@@ -3379,7 +3379,7 @@ object GpuOverrides extends Logging {
             willNotWorkOnGpu("Fail on error is not supported on GPU when parsing urls.")
           }
 
-          extractStringLit(a.children(1)).map(_.toUpperCase) match {
+          extractStringLit(a.children(1)) match {
             // In Spark, the key in parse_url could act like a regex, but GPU will match the key
             // exactly. When key is literal, GPU will check if the key contains regex special and
             // fallbcak to CPU if it does, but we are not able to fallback when key is column.

sql-plugin/src/main/scala/org/apache/spark/sql/rapids/GpuParseUrl.scala

@@ -40,10 +40,10 @@ object GpuParseUrl {
 
   def isSupportedPart(part: String): Boolean = {
     part match {
-      case PROTOCOL | HOST | QUERY | PATH =>
-        true
-      case _ =>
+      case REF | FILE | AUTHORITY | USERINFO =>
         false
+      case _ => // PROTOCOL, HOST, QUERY, PATH and invalid parts are supported
+        true
     }
   }
 }
@@ -73,7 +73,7 @@ case class GpuParseUrl(children: Seq[Expression])
         throw new UnsupportedOperationException(s"$this is not supported partToExtract=$part. " +
             s"Only PROTOCOL, HOST, QUERY and PATH are supported")
       case _ =>
-        throw new IllegalArgumentException(s"Invalid partToExtract: $partToExtract")
+        return GpuColumnVector.columnVectorFromNull(url.getRowCount.toInt, StringType)
     }
   }