Compile Flex lexer with options for better code generation

2 files changed, 17 insertions, 0 deletions

diff --git a/document/src/tests/documentselectparsertest.cpp b/document/src/tests/documentselectparsertest.cpp
index c8cbd1be9c2..fce0d90db72 100644
--- a/document/src/tests/documentselectparsertest.cpp
+++ b/document/src/tests/documentselectparsertest.cpp
@@ -1653,4 +1653,14 @@ TEST_F(DocumentSelectParserTest, selection_has_upper_limit_on_input_size) {
     verifyFailedParse(expr, "ParsingFailedException: expression is too large to be parsed");
 }
 
+TEST_F(DocumentSelectParserTest, lexing_does_not_have_superlinear_time_complexity) {
+    createDocs();
+    std::string expr = ("testdoctype1.hstringval == 'a_biii"
+                        + std::string(select::ParserLimits::MaxSelectionByteSize - 100, 'i')
+                        + "iiig string'");
+    // If the lexer is not compiled with the appropriate options, this will take a long time.
+    // A really, really long time.
+    PARSE(expr, *_doc[0], False);
+}
+
 } // document
diff --git a/document/src/vespa/document/select/grammar/lexer.ll b/document/src/vespa/document/select/grammar/lexer.ll
index bd011c8ebf6..1222aac02a2 100644
--- a/document/src/vespa/document/select/grammar/lexer.ll
+++ b/document/src/vespa/document/select/grammar/lexer.ll
@@ -7,6 +7,13 @@
 %option noyywrap nounput
 %option yyclass="document::select::DocSelScanner"
 
+ /* Flex lexer must be compiled with batch mode (as opposed to interactive mode)
+  * or parsing of large tokens appears to trigger superlinear time complexity.
+  * Also use full, non-compressed lookup tables for maximum performance.
+  */
+%option batch
+%option full
+
  /* Used to track source locations, see https://github.com/bingmann/flex-bison-cpp-example/blob/master/src/scanner.ll */
 %{
 #define YY_USER_ACTION yyloc->columns(yyleng);