From 589923e168d45c76a230227c1a9dfebd2f5b6990 Mon Sep 17 00:00:00 2001
From: Arne Juul <arnej@verizonmedia.com>
Date: Tue, 13 Jul 2021 11:45:59 +0000
Subject: try to trap spurious failure

* we have seen spurious failures when verifying output from accent dropping;
  so far nothing reproducible, so add some extra logging and retry once
  if it happens (in case it's some kind of race-condition glitch).
---
 .../expressions/NormalizeExpression.java           | 30 +++++++++++++++++++
 .../expressions/NormalizeTestCase.java             | 35 ++++++++++++++++++++++
 2 files changed, 65 insertions(+)

diff --git a/indexinglanguage/src/main/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeExpression.java b/indexinglanguage/src/main/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeExpression.java
index 9c6e698a154..78a7a858aba 100644
--- a/indexinglanguage/src/main/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeExpression.java
+++ b/indexinglanguage/src/main/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeExpression.java
@@ -6,12 +6,16 @@ import com.yahoo.document.datatypes.StringFieldValue;
 import com.yahoo.language.Linguistics;
 import com.yahoo.language.process.Transformer;
 
+import java.util.logging.Logger;
+import java.util.logging.Level;
+
 /**
  * @author Simon Thoresen Hult
  */
 public final class NormalizeExpression extends Expression {
 
     private final Linguistics linguistics;
+    private static final Logger logger = Logger.getLogger(NormalizeExpression.class.getName());
 
     public NormalizeExpression(Linguistics linguistics) {
         super(DataType.STRING);
@@ -22,9 +26,35 @@ public final class NormalizeExpression extends Expression {
         return linguistics;
     }
 
+    
+    private static String escape(String str) {
+        StringBuilder buf = new StringBuilder();
+        for (char c : str.toCharArray()) {
+            if (c >= ' ') {
+                buf.append(c);
+            } else {
+                buf.append(String.format("U+%04X", (int)c));
+            }
+        }
+        return buf.toString();
+    }
+
     @Override
     protected void doExecute(ExecutionContext context) {
         Transformer transformer = linguistics.getTransformer();
+        var orig = String.valueOf(context.getValue());
+        var lang = context.resolveLanguage(linguistics);
+        var transformed = transformer.accentDrop(orig, lang);
+        try {
+            context.setValue(new StringFieldValue(transformed));
+            return;
+        } catch (IllegalArgumentException ex) {
+            String msg = ("bad normalize, \n" +
+                          "original: >>> " + escape(orig) + " <<<\n" +
+                          " -> accentDrop(" + lang + ") -> \n" +
+                          "transformed: >>> " + escape(transformed) + " <<<");
+            logger.log(Level.SEVERE, msg);
+        }
         context.setValue(new StringFieldValue(transformer.accentDrop(String.valueOf(context.getValue()),
                                                                      context.resolveLanguage(linguistics))));
     }
diff --git a/indexinglanguage/src/test/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeTestCase.java b/indexinglanguage/src/test/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeTestCase.java
index 8b4f1a8b344..3ad1b129b4d 100644
--- a/indexinglanguage/src/test/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeTestCase.java
+++ b/indexinglanguage/src/test/java/com/yahoo/vespa/indexinglanguage/expressions/NormalizeTestCase.java
@@ -6,8 +6,10 @@ import com.yahoo.document.datatypes.FieldValue;
 import com.yahoo.document.datatypes.StringFieldValue;
 import com.yahoo.language.Language;
 import com.yahoo.language.Linguistics;
+import com.yahoo.language.process.Transformer;
 import com.yahoo.language.simple.SimpleLinguistics;
 import com.yahoo.vespa.indexinglanguage.SimpleTestAdapter;
+
 import org.junit.Test;
 import org.mockito.Mockito;
 
@@ -56,4 +58,37 @@ public class NormalizeTestCase {
         assertTrue(val instanceof StringFieldValue);
         assertEquals("beyonce", ((StringFieldValue)val).getString());
     }
+
+    class MyMockTransformer implements Transformer {
+        boolean first = true;
+        @Override
+        public String accentDrop(String input, Language language) {
+            if (first) {
+                first = false;
+                return input.replace(' ', '\u0008');
+            } else {
+                return input.replace(' ', '/');
+            }
+        }
+    }        
+
+    class MyMockLinguistics extends SimpleLinguistics {
+        private Transformer transformer = new MyMockTransformer();
+        @Override
+        public Transformer getTransformer() {
+            return transformer;
+        }
+    }
+
+    @Test
+    public void requireThatBadNormalizeRetries() {
+        ExecutionContext ctx = new ExecutionContext(new SimpleTestAdapter());
+        ctx.setLanguage(Language.ENGLISH);
+        ctx.setValue(new StringFieldValue("bad norm"));
+        var linguistics = new MyMockLinguistics();
+        new NormalizeExpression(linguistics).execute(ctx);
+        FieldValue val = ctx.getValue();
+        assertTrue(val instanceof StringFieldValue);
+        assertEquals("bad/norm", ((StringFieldValue)val).getString());
+    }
 }
-- 
cgit v1.2.3