「優化」獲得光標在安卓VS列名的位置

Question

從性能上看：是不是很好，如果在每次訪問我的遊標我使用類似：

public static final String COLUMN_NAME = "my_column_name"; 
cursor.getString(cursor.getColumnIndex(COLUMN_NAME)); 

或者我應該看衡量提高性能，如果我用這個來代替：

public static final int COLUMN_POSITION = #column_position; 
cursor.getString(COLUMN_POSITION); 

我更喜歡第一種方法，因爲其餘代碼不依賴於在查詢中列的位置，但只的名字柱。所以我的問題是：爲了改善使用常量位置訪問遊標的性能，值得犧牲這一點？你喜歡什麼方式在你的Android應用程序？

Answer 1

我想這與android無關。在SQLite中，通過索引訪問列（列位置）應該更快。

Answer 2

要回答你的問題（和我的方式），我做了一些測試。

這次試驗主要檢查多少時間查詢了那些兩種情況：

1. 使用cursor.getString(cursor.getColumnIndex(COLUMN_NAME))方法
2. 第一獲取列ID，然後直接調用cursor.getString(COLUMN_POSITION)方法

爲了使性能測試具有實際意義，我在數據庫中插入了行，然後我做了一個查詢，投擲我的ContentProvider這些元素。

結果：

___________________________________________________________________________ 
| Column count| Time (ms) getColumnIndex | Time (ms) columnId | improvement | 
|_____________|__________________________|____________________|_____________| 
| 500   | 34564     | 30401    | 13%   | 
| 200   | 9987     | 8502    | 17%   | 
| 100   | 4713     | 4004    | 17%   | 
| 50   | 2400     | 1971    | 21%   | 
| 20   | 1088     | 915    | 19%   | 
|___________________________________________________________________________| 

因此，首先讓列ID和呼叫直接getString()方法將花費大約20％更少的時間。

---

試驗方法的細節：

平臺：在Android 4.3 的Nexus 7（2012）

數據庫創建：

public static int TESTSPEEDCOLUMNCOUNT = 200; 
StringBuilder sb = new StringBuilder(); 
sb.append("CREATE TABLE " + Tables.TESTSPEED + " ("); 
sb.append(BaseColumns._ID + " INTEGER PRIMARY KEY AUTOINCREMENT, "); 
for (int i = 0; i < (TESTSPEEDCOLUMNCOUNT - 1); ++i) { 
    sb.append("C" + i + " TEXT, "); 
} 
sb.append("C" + (TESTSPEEDCOLUMNCOUNT - 1) + " TEXT)"); 
db.execSQL(sb.toString()); 

測試用例：

public class ProviderTestSpeed extends ProviderTestCase2<MyProvider> { 

    private ContentValues createElementForId(String id) { 
     ContentValues cv = new ContentValues(); 
     for (int i = 0; i < TESTSPEEDCOLUMNCOUNT; ++i) { 
      cv.put("C" + i, id); 
     } 
     return cv; 
    } 



    public void testSpeed() { 
     Log.d(TAG, "testSpeed start columnCount = " + columnCount); 
     ArrayList<ContentValues> list = new ArrayList<ContentValues>(); 
     ContentValues[] tabcv = {}; 
     for (int j = 0; j < 10; ++j) { 
      list.clear(); 
      for (int i = 0; i < 500; ++i) { 
       ContentValues cv = createElementForId(String.valueOf(i)); 
       list.add(cv); 
      } 
      mContentResolver.bulkInsert(TestSpeedCONTENT_URI, list.toArray(tabcv)); 
     } 
     Log.d(TAG, "testSpeed insertFinished"); 
     Cursor cursor = mContentResolver.query(TestSpeedCONTENT_URI, null, null, null, null); 
     cursor.moveToFirst(); 
     Log.d(TAG, "testSpeed itemCount = " + cursor.getCount() + " columnCount=" + cursor.getColumnCount()); 

     // build the tab to avoid dynamic allocation during the mesure 
     ArrayList<String> listColumns = new ArrayList<String>(); 
     for (int i = 0; i < TESTSPEEDCOLUMNCOUNT; ++i) { 
      listColumns.add("C" + i); 
     } 
     String[] tabColumnsType = {}; 
     String[] tabColumns = listColumns.toArray(tabColumnsType); 

     Date now = new Date(); 
     long start = now.getTime(); 
     do { 
      for (int i = 0; i < TESTSPEEDCOLUMNCOUNT; ++i) { 
       // get the all the columns of the table 
       cursor.getString(cursor.getColumnIndex(tabColumns[i])); 
      } 
     } while (cursor.moveToNext()); 
     now = new Date(); 
     long end = now.getTime(); 

     Log.d(TAG, "testSpeed took " + (end - start) + " with getColumnIndex at each time"); 
     cursor.moveToFirst(); 
     now = new Date(); 
     start = now.getTime(); 
     do { 
      for (int i = 0; i < TESTSPEEDCOLUMNCOUNT; ++i) { 
       // get the all the columns of the table using directly the column id 
       cursor.getString(i); 
      } 
     } while (cursor.moveToNext()); 
     now = new Date(); 
     end = now.getTime(); 
     Log.d(TAG, "testSpeed took " + (end - start) + " with getColumnIndex before loop"); 
    } 
} 

我認爲200到500之間的性能下降來自光標窗口。我有很多像200列以上的日誌：

W/CursorWindow(1628): Window is full: requested allocation 2412 bytes, free space 988 bytes, window size 2097152 bytes