Fix TimeZone.getAvailableIDs(int).

This was broken by the removal of the pre-computed raw offsets from
the tzdata file. I think that's still the direction we want to go (with
us hopefully using more of icu4j at some point, and eventually relying
solely on the icu time zone data), so this patch adds code to lazily
evaluate all the offsets by instantiating all the time zones.

Bug: 16947622
Change-Id: I6d1dfe5ee6c99338f9807c3af5b6f04539c256c3

1 files changed, 24 insertions, 8 deletions

diff --git a/luni/src/main/java/libcore/util/ZoneInfoDB.java b/luni/src/main/java/libcore/util/ZoneInfoDB.java
index 07aaf04..a9d06a4 100644
--- a/luni/src/main/java/libcore/util/ZoneInfoDB.java
+++ b/luni/src/main/java/libcore/util/ZoneInfoDB.java
@@ -62,11 +62,11 @@ public final class ZoneInfoDB {
     /**
      * The 'ids' array contains time zone ids sorted alphabetically, for binary searching.
      * The other two arrays are in the same order. 'byteOffsets' gives the byte offset
-     * of each time zone, and 'rawUtcOffsets' gives the time zone's raw UTC offset.
+     * of each time zone, and 'rawUtcOffsetsCache' gives the time zone's raw UTC offset.
      */
     private String[] ids;
     private int[] byteOffsets;
-    private int[] rawUtcOffsets;
+    private int[] rawUtcOffsetsCache; // Access this via getRawUtcOffsets instead.
 
     /**
      * ZoneInfo objects are worth caching because they are expensive to create.
@@ -104,7 +104,7 @@ public final class ZoneInfoDB {
       version = "missing";
       zoneTab = "# Emergency fallback data.\n";
       ids = new String[] { "GMT" };
-      byteOffsets = rawUtcOffsets = new int[1];
+      byteOffsets = rawUtcOffsetsCache = new int[1];
     }
 
     private boolean loadData(String path) {
@@ -171,7 +171,6 @@ public final class ZoneInfoDB {
       int idOffset = 0;
 
       byteOffsets = new int[entryCount];
-      rawUtcOffsets = new int[entryCount];
 
       for (int i = 0; i < entryCount; i++) {
         it.readByteArray(idBytes, 0, idBytes.length);
@@ -183,7 +182,7 @@ public final class ZoneInfoDB {
         if (length < 44) {
           throw new AssertionError("length in index file < sizeof(tzhead)");
         }
-        rawUtcOffsets[i] = it.readInt();
+        it.skip(4); // Skip the unused 4 bytes that used to be the raw offset.
 
         // Don't include null chars in the String
         int len = idBytes.length;
@@ -210,16 +209,33 @@ public final class ZoneInfoDB {
       return ids.clone();
     }
 
-    public String[] getAvailableIDs(int rawOffset) {
+    public String[] getAvailableIDs(int rawUtcOffset) {
       List<String> matches = new ArrayList<String>();
-      for (int i = 0, end = rawUtcOffsets.length; i < end; ++i) {
-        if (rawUtcOffsets[i] == rawOffset) {
+      int[] rawUtcOffsets = getRawUtcOffsets();
+      for (int i = 0; i < rawUtcOffsets.length; ++i) {
+        if (rawUtcOffsets[i] == rawUtcOffset) {
           matches.add(ids[i]);
         }
       }
       return matches.toArray(new String[matches.size()]);
     }
 
+    private synchronized int[] getRawUtcOffsets() {
+      if (rawUtcOffsetsCache != null) {
+        return rawUtcOffsetsCache;
+      }
+      rawUtcOffsetsCache = new int[ids.length];
+      for (int i = 0; i < ids.length; ++i) {
+        // This creates a TimeZone, which is quite expensive. Hence the cache.
+        // Note that icu4c does the same (without the cache), so if you're
+        // switching this code over to icu4j you should check its performance.
+        // Telephony shouldn't care, but someone converting a bunch of calendar
+        // events might.
+        rawUtcOffsetsCache[i] = cache.get(ids[i]).getRawOffset();
+      }
+      return rawUtcOffsetsCache;
+    }
+
     public String getVersion() {
       return version;
     }