ICU-23321 Eliminate lock contention and fix concurrency bugs across ICU4J

icu4j/main/charset/src/main/java/com/ibm/icu/charset/CharsetISCII.java

@@ -51,8 +51,34 @@ class CharsetISCII extends CharsetICU {
     private static final int NO_CHAR_MARKER = 0xfffe;
 
     /* Used for proper conversion to and from Gurmukhi */
-    private static UnicodeSet PNJ_BINDI_TIPPI_SET;
-    private static UnicodeSet PNJ_CONSONANT_SET;
+    private static final class PNJSetsHolder {
+        static final UnicodeSet PNJ_BINDI_TIPPI_SET;
+        static final UnicodeSet PNJ_CONSONANT_SET;
+
+        static {
+            UnicodeSet consonant = new UnicodeSet();
+            consonant.add(0x0a15, 0x0a28);
+            consonant.add(0x0a2a, 0x0a30);
+            consonant.add(0x0a35, 0x0a36);
+            consonant.add(0x0a38, 0x0a39);
+
+            UnicodeSet bindiTippi = new UnicodeSet();
+            bindiTippi.addAll(consonant);
+            bindiTippi.add(0x0a05);
+            bindiTippi.add(0x0a07);
+            bindiTippi.add(0x0a41, 0x0a42);
+            bindiTippi.add(0x0a3f);
+
+            consonant.compact();
+            bindiTippi.compact();
+            consonant.freeze();
+            bindiTippi.freeze();
+
+            PNJ_CONSONANT_SET = consonant;
+            PNJ_BINDI_TIPPI_SET = bindiTippi;
+        }
+    }
+
     private static final short PNJ_BINDI = 0x0a02;
     private static final short PNJ_TIPPI = 0x0a70;
     private static final short PNJ_SIGN_VIRAMA = 0x0a4d;
@@ -1285,32 +1311,6 @@ public CharsetISCII(String icuCanonicalName, String javaCanonicalName, String[]
         int option = Integer.parseInt(icuCanonicalName.substring(14));
 
         extraInfo = new UConverterDataISCII(option);
-
-        initializePNJSets();
-    }
-
-    /* Initialize the two UnicodeSets use for proper Gurmukhi conversion if they have not already been created. */
-    private void initializePNJSets() {
-        if (PNJ_BINDI_TIPPI_SET != null && PNJ_CONSONANT_SET != null) {
-            return;
-        }
-        PNJ_BINDI_TIPPI_SET = new UnicodeSet();
-        PNJ_CONSONANT_SET = new UnicodeSet();
-
-        PNJ_CONSONANT_SET.add(0x0a15, 0x0a28);
-        PNJ_CONSONANT_SET.add(0x0a2a, 0x0a30);
-        PNJ_CONSONANT_SET.add(0x0a35, 0x0a36);
-        PNJ_CONSONANT_SET.add(0x0a38, 0x0a39);
-
-        PNJ_BINDI_TIPPI_SET.addAll(PNJ_CONSONANT_SET);
-        PNJ_BINDI_TIPPI_SET.add(0x0a05);
-        PNJ_BINDI_TIPPI_SET.add(0x0a07);
-
-        PNJ_BINDI_TIPPI_SET.add(0x0a41, 0x0a42);
-        PNJ_BINDI_TIPPI_SET.add(0x0a3f);
-
-        PNJ_CONSONANT_SET.compact();
-        PNJ_BINDI_TIPPI_SET.compact();
     }
 
     /*
@@ -1691,7 +1691,8 @@ protected CoderResult decodeLoop(
                         /* Check to make sure that consonant clusters are handled correctly for Gurmukhi script. */
                         if (data.currentDeltaToUnicode == PNJ_DELTA
                                 && data.prevToUnicodeStatus != 0
-                                && PNJ_CONSONANT_SET.contains(data.prevToUnicodeStatus)
+                                && PNJSetsHolder.PNJ_CONSONANT_SET.contains(
+                                        data.prevToUnicodeStatus)
                                 && (this.toUnicodeStatus + PNJ_DELTA) == PNJ_SIGN_VIRAMA
                                 && (targetUniChar + PNJ_DELTA) == data.prevToUnicodeStatus) {
                             if (offsets != null) {
@@ -1730,7 +1731,7 @@ protected CoderResult decodeLoop(
                              */
                             if (data.currentDeltaToUnicode == PNJ_DELTA
                                     && (targetUniChar + PNJ_DELTA) == PNJ_BINDI
-                                    && PNJ_BINDI_TIPPI_SET.contains(
+                                    && PNJSetsHolder.PNJ_BINDI_TIPPI_SET.contains(
                                             this.toUnicodeStatus + PNJ_DELTA)) {
                                 targetUniChar = PNJ_TIPPI - PNJ_DELTA;
                                 cr =
@@ -1743,7 +1744,7 @@ protected CoderResult decodeLoop(
                                                 PNJ_DELTA);
                             } else if (data.currentDeltaToUnicode == PNJ_DELTA
                                     && (targetUniChar + PNJ_DELTA) == PNJ_SIGN_VIRAMA
-                                    && PNJ_CONSONANT_SET.contains(
+                                    && PNJSetsHolder.PNJ_CONSONANT_SET.contains(
                                             this.toUnicodeStatus + PNJ_DELTA)) {
                                 /* Store the current toUnicodeStatus code point for later handling of consonant cluster in Gurmukhi. */
                                 data.prevToUnicodeStatus = this.toUnicodeStatus + PNJ_DELTA;
@@ -2041,7 +2042,7 @@ protected CoderResult encodeLoop(
                 } // end of switch
                 if (converterData.currentDeltaFromUnicode == PNJ_DELTA
                         && tempContextFromUnicode == PNJ_ADHAK
-                        && PNJ_CONSONANT_SET.contains(sourceChar + PNJ_DELTA)) {
+                        && PNJSetsHolder.PNJ_CONSONANT_SET.contains(sourceChar + PNJ_DELTA)) {
                     /* If the previous codepoint is Adhak and the current codepoint is a consonant, the targetByteUnit should be C + Halant + C. */
                     /* reset context char */
                     converterData.contextCharFromUnicode = 0x0000;

icu4j/main/charset/src/main/java/com/ibm/icu/charset/CharsetProviderICU.java

@@ -13,7 +13,6 @@
 import java.nio.charset.Charset;
 import java.nio.charset.UnsupportedCharsetException;
 import java.nio.charset.spi.CharsetProvider;
-import java.util.Collections;
 import java.util.Iterator;
 import java.util.LinkedList;
 import java.util.List;
@@ -25,11 +24,31 @@
  */
 public final class CharsetProviderICU extends CharsetProvider {
     /**
-     * List of available ICU Charsets, empty during static initialization. Not a Set or Map, so that
-     * we can add different Charset objects with the same name(), which means that they are
-     * .equals(). See ICU ticket #11493.
+     * Holds the list of available ICU Charsets, loaded once. Uses the holder class idiom (JLS
+     * 12.4.2) for thread-safe lazy initialization. Not a Set or Map, so that we can add different
+     * Charset objects with the same name(), which means that they are .equals(). See ICU ticket
+     * #11493.
      */
-    private static List<Charset> icuCharsets = Collections.<Charset>emptyList();
+    private static final class AvailableCharsetsHolder {
+        static final List<Charset> CHARSETS;
+
+        static {
+            List<Charset> icucs = new LinkedList<Charset>();
+            int num = UConverterAlias.countAvailable();
+            for (int i = 0; i < num; ++i) {
+                String name = UConverterAlias.getAvailableName(i);
+                try {
+                    Charset cs = getCharset(name, "");
+                    icucs.add(cs);
+                } catch (UnsupportedCharsetException ex) {
+                } catch (IOException e) {
+                }
+                // add only charsets that can be created!
+            }
+            // Unmodifiable so that charsets().next().remove() cannot change it.
+            CHARSETS = List.copyOf(icucs);
+        }
+    }
 
     /**
      * Default constructor
@@ -264,28 +283,8 @@ private static final String[] getAliases(String encName) throws IOException {
         return (ret);
     }
 
-    /**
-     * Lazy-init the icuCharsets list. Could be done during static initialization if constructing
-     * all of the Charsets were cheap enough. See ICU ticket #11481.
-     */
-    private static final synchronized void loadAvailableICUCharsets() {
-        if (!icuCharsets.isEmpty()) {
-            return;
-        }
-        List<Charset> icucs = new LinkedList<Charset>();
-        int num = UConverterAlias.countAvailable();
-        for (int i = 0; i < num; ++i) {
-            String name = UConverterAlias.getAvailableName(i);
-            try {
-                Charset cs = getCharset(name, "");
-                icucs.add(cs);
-            } catch (UnsupportedCharsetException ex) {
-            } catch (IOException e) {
-            }
-            // add only charsets that can be created!
-        }
-        // Unmodifiable so that charsets().next().remove() cannot change it.
-        icuCharsets = Collections.unmodifiableList(icucs);
+    private static List<Charset> getAvailableICUCharsets() {
+        return AvailableCharsetsHolder.CHARSETS;
     }
 
     /**
@@ -297,8 +296,7 @@ private static final synchronized void loadAvailableICUCharsets() {
      */
     @Override
     public final Iterator<Charset> charsets() {
-        loadAvailableICUCharsets();
-        return icuCharsets.iterator();
+        return getAvailableICUCharsets().iterator();
     }
 
     /**
@@ -310,10 +308,10 @@ public final Iterator<Charset> charsets() {
      */
     @Deprecated
     public static final String[] getAvailableNames() {
-        loadAvailableICUCharsets();
-        String[] names = new String[icuCharsets.size()];
+        List<Charset> charsets = getAvailableICUCharsets();
+        String[] names = new String[charsets.size()];
         int i = 0;
-        for (Charset cs : icuCharsets) {
+        for (Charset cs : charsets) {
             names[i++] = cs.name();
         }
         return names;

icu4j/main/charset/src/main/java/com/ibm/icu/charset/UConverterAlias.java

@@ -100,7 +100,7 @@ private static final int strlen(byte[] sArray, int sBegin) {
                                                                  * offsets[]
                                                                  */
 
-    static ByteBuffer gAliasData = null;
+    static volatile ByteBuffer gAliasData = null;
 
     private static final boolean isAlias(String alias) {
         if (alias == null) {
@@ -111,14 +111,16 @@ private static final boolean isAlias(String alias) {
 
     private static final String CNVALIAS_DATA_FILE_NAME = "cnvalias.icu";
 
-    private static final synchronized boolean haveAliasData() throws IOException {
-        boolean needInit;
-
-        needInit = gAliasData == null;
+    private static boolean haveAliasData() throws IOException {
+        if (gAliasData != null) {
+            return true;
+        }
+        synchronized (UConverterAlias.class) {
+            if (gAliasData != null) {
+                return true;
+            }
 
-        /* load converter alias data from file if necessary */
-        if (needInit) {
-            ByteBuffer data = null;
+            /* load converter alias data from file if necessary */
             int[] tableArray = null;
             int tableStart;
 
@@ -142,17 +144,12 @@ private static final synchronized boolean haveAliasData() throws IOException {
             gNormalizedStringTable = new byte[tableArray[normalizedStringTableIndex] * 2];
             b.get(gNormalizedStringTable);
 
-            data = ByteBuffer.allocate(0); // dummy UDataMemory object in absence
-            // of memory mapping
-
             if (gOptionTable[0] != STD_NORMALIZED) {
                 throw new IOException("Unsupported alias normalization");
             }
 
-            if (gAliasData == null) {
-                gAliasData = data;
-                data = null;
-            }
+            // Write gAliasData last: it's the volatile flag read by the fast path
+            gAliasData = ByteBuffer.allocate(0);
         }
 
         return true;

icu4j/main/collate/src/main/java/com/ibm/icu/impl/coll/CollationTailoring.java

@@ -119,7 +119,7 @@ int getUCAVersion() {
     CollationData ownedData;
     Trie2_32 trie;
     UnicodeSet unsafeBackwardSet;
-    public Map<Integer, Integer> maxExpansions;
+    public volatile Map<Integer, Integer> maxExpansions;
 
     /*
      * Not Cloneable: A CollationTailoring cannot be copied.

icu4j/main/collate/src/main/java/com/ibm/icu/impl/text/RbnfScannerProviderImpl.java

@@ -16,8 +16,7 @@
 import com.ibm.icu.text.RbnfLenientScannerProvider;
 import com.ibm.icu.text.RuleBasedCollator;
 import com.ibm.icu.util.ULocale;
-import java.util.HashMap;
-import java.util.Map;
+import java.util.concurrent.ConcurrentHashMap;
 
 /**
  * Returns RbnfLenientScanners that use the old RuleBasedNumberFormat implementation behind
@@ -29,15 +28,15 @@
 @Deprecated
 public class RbnfScannerProviderImpl implements RbnfLenientScannerProvider {
     private static final boolean DEBUG = ICUDebug.enabled("rbnf");
-    private Map<String, RbnfLenientScanner> cache;
+    private final ConcurrentHashMap<String, RbnfLenientScanner> cache;
 
     /**
      * @internal
      * @deprecated This API is ICU internal only.
      */
     @Deprecated
     public RbnfScannerProviderImpl() {
-        cache = new HashMap<String, RbnfLenientScanner>();
+        cache = new ConcurrentHashMap<>();
     }
 
     /**
@@ -73,17 +72,16 @@ public RbnfScannerProviderImpl() {
     public RbnfLenientScanner get(ULocale locale, String extras) {
         RbnfLenientScanner result = null;
         String key = locale.toString() + "/" + extras;
-        synchronized (cache) {
-            result = cache.get(key);
-            if (result != null) {
-                return result;
-            }
+        // Avoid computeIfAbsent: createScanner() constructs a RuleBasedCollator (expensive)
+        // and would hold the ConcurrentHashMap bin lock for the entire duration. Using
+        // get-then-putIfAbsent allows parallel construction with first-write-wins.
+        result = cache.get(key);
+        if (result != null) {
+            return result;
         }
         result = createScanner(locale, extras);
-        synchronized (cache) {
-            cache.put(key, result);
-        }
-        return result;
+        RbnfLenientScanner existing = cache.putIfAbsent(key, result);
+        return existing != null ? existing : result;
     }
 
     /**

icu4j/main/collate/src/main/java/com/ibm/icu/text/RuleBasedCollator.java

@@ -264,6 +264,9 @@ public RuleBasedCollator clone() throws CloneNotSupportedException {
     }
 
     private final void initMaxExpansions() {
+        if (tailoring.maxExpansions != null) {
+            return;
+        }
         synchronized (tailoring) {
             if (tailoring.maxExpansions == null) {
                 tailoring.maxExpansions =

icu4j/main/collate/src/test/java/com/ibm/icu/dev/test/util/CollationConcurrencyTest.java

@@ -0,0 +1,42 @@
+// © 2026 and later: Unicode, Inc. and others.
+// License & terms of use: http://www.unicode.org/copyright.html
+package com.ibm.icu.dev.test.util;
+
+import com.ibm.icu.text.CollationElementIterator;
+import com.ibm.icu.text.Collator;
+import com.ibm.icu.text.RuleBasedCollator;
+import com.ibm.icu.util.ULocale;
+import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.JUnit4;
+
+/** Concurrency regression tests for CollationTailoring. */
+@RunWith(JUnit4.class)
+public class CollationConcurrencyTest extends ConcurrencyTest {
+
+    /** Concurrent getMaxExpansion() must see the volatile maxExpansions field. */
+    @Test
+    public void testCollationMaxExpansionsConcurrent() throws Exception {
+        String[] localeIDs = {"en", "de", "ja", "zh", "ko", "fr"};
+        runConcurrent(
+                "CollationMaxExpansions",
+                tid -> {
+                    for (int i = 0; i < ITERATIONS; i++) {
+                        String localeID = localeIDs[(tid + i) % localeIDs.length];
+                        RuleBasedCollator col =
+                                (RuleBasedCollator) Collator.getInstance(new ULocale(localeID));
+                        assertNotNull("Collator should not be null for " + localeID, col);
+                        // getCollationElementIterator triggers initMaxExpansions()
+                        // which reads/writes the volatile maxExpansions field
+                        CollationElementIterator cei =
+                                col.getCollationElementIterator("test string");
+                        int ce = cei.next();
+                        assertTrue(
+                                "should produce at least one collation element",
+                                ce != CollationElementIterator.NULLORDER);
+                        int maxExp = cei.getMaxExpansion(ce);
+                        assertTrue("max expansion should be positive", maxExp > 0);
+                    }
+                });
+    }
+}