// org.springframework.jdbc.core.JdbcTemplate 中的查詢方法基本都有支持參數RowMapper<T> rowMapper的重載方法。下面只是隨便舉例2個,還有很多 public <T> List<T> query(String sql, Object[] args, RowMapper<T> rowMapper) throws DataAccessException { ... }; public <T> T queryForObject(String sql, Object[] args, RowMapper<T> rowMapper) throws DataAccessException { ... }; //demo01 List<Person> person = jdbcTemplate.query(sql, new RowMapper<Person>() { @Override public Person mapRow(ResultSet rs, int i) throws SQLException { Person p = new Person(); //特別需要new,不然如果結果集是list就只有1個對象 p.setId(rs.getString("id")); return p; }}); //特別 如果如demo寫,很麻煩要set很多。此時spring提供了一個RowMapper的實現類BeanPropertyRowMapper //demo02 List<Person> person = jdbcTemplate.query(sql, new BeanPropertyRowMapper(Person.class));
這篇博客的主要目的是分析BeanPropertyRowMapper的實現是怎么樣。
先,之前也在用jdbcTemplate來查詢,但都是用demo01的方式。上周末本來想寫一個BaseRowMapper(其實就是BeanPropertyRowMapper),但冬(lan)眠(si)去了。
在今天(2016-11-07)上班的時候又用到了,於是就打算寫。但看了下RowMapper的結構(ctrl+t)發現了BeanPropertyRowMapper。這不就是我想要的嗎,於是決定去看下和自己的想法有什么差別。
現在,我還沒看過源碼,先說我周末大致想到的:
1、肯定要用反射,根據sql的列名/別名去找到對應的set;
以sql返回結果集的列為准,sql有的必須有set,有set不一定sql有返回。
2、反射效率低,如果我sql返回的是list,不應該每行都要根據反射去找set。而是應該在第一次找的時候,把列名/別名對應的set緩存起來,以后直接取;
第一次(第一行結果)用列名/別名(不區分大小寫),找到set,並緩存;之后直接用別名/列名去找set,節約反射查找set消耗的時間。
注:以下都是BeanPropertyRowMapper源碼分析
一、緩存自定義類型的set方法
// BeanPropertyRowMapper的成員變量 /** Logger available to subclasses */ protected final Log logger = LogFactory.getLog(getClass()); /** The class we are mapping to ;要映射的class*/ private Class<T> mappedClass; /** Whether we're strictly validating; 是否嚴格映射bean和sql結果 */ private boolean checkFullyPopulated = false; /** Whether we're defaulting primitives when mapping a null value */ private boolean primitivesDefaultedForNullValue = false; /** Map of the fields we provide mapping for;映射字段的set方法 */ private Map<String, PropertyDescriptor> mappedFields; /** Set of bean properties we provide mapping for ;需要映射的字段*/ private Set<String> mappedProperties;
/** * Create a new BeanPropertyRowMapper, accepting unpopulated properties in the target bean. * <p>Consider using the {@link #newInstance} factory method instead,which allows for specifying the mapped type once only. * @param mappedClass the class that each row should be mapped to */ public BeanPropertyRowMapper(Class<T> mappedClass) { initialize(mappedClass); } /** * Create a new BeanPropertyRowMapper. * @param mappedClass the class that each row should be mapped to * @param checkFullyPopulated whether we're strictly validating that all bean properties have been mapped from corresponding database fields */ public BeanPropertyRowMapper(Class<T> mappedClass, boolean checkFullyPopulated) { initialize(mappedClass); this.checkFullyPopulated = checkFullyPopulated; //是否嚴格驗證,所有bean屬性已經從對應的數據庫字段映射。 }
在BeanPropertyRowMapper提供的2中構造函數中,區別只在於是否嚴格映射bean和sql結果(默認是false,不嚴格映射)。
/** * Initialize the mapping metadata for the given class. * @param mappedClass the mapped class. */ protected void initialize(Class<T> mappedClass) { this.mappedClass = mappedClass; this.mappedFields = new HashMap<String, PropertyDescriptor>(); this.mappedProperties = new HashSet<String>(); //以上都是設置/初始化成員變量 PropertyDescriptor[] pds = BeanUtils.getPropertyDescriptors(mappedClass);//org.springframework.beans.BeanUtils
for (PropertyDescriptor pd : pds) { if (pd.getWriteMethod() != null) { this.mappedFields.put(pd.getName().toLowerCase(), pd); // key:全小寫 String underscoredName = underscoreName(pd.getName()); // ex:bookName --> book_name if (!pd.getName().toLowerCase().equals(underscoredName)) { //set與其屬性命名的不一致;方法是setBookName 而變量是book_name; 大致是這意思 this.mappedFields.put(underscoredName, pd); } this.mappedProperties.add(pd.getName()); //key:與mappedFields不一樣 } } } /** * Convert a name in camelCase to an underscored name in lower case. * Any upper case letters are converted to lower case with a preceding underscore. * @param name the string containing original name * @return the converted name */ private String underscoreName(String name) { //ex: bookName --> book_name if (!StringUtils.hasLength(name)) { return ""; } StringBuilder result = new StringBuilder(); result.append(name.substring(0, 1).toLowerCase()); for (int i = 1; i < name.length(); i++) { String s = name.substring(i, i + 1); String slc = s.toLowerCase(); if (!s.equals(slc)) { //大寫字母轉換成 _+小寫 result.append("_").append(slc); } else { result.append(s); } } return result.toString(); }
注意:
1、Map<String, PropertyDescriptor> mappedFields的key與Set<String> mappedProperties的value保存的並不一定是一樣的:
mappedFields的key是set方法的全小寫/帶下划線的全小寫,而mappedProperties的是set方法名。
ex: private String bookName; public void setBookName(..)
mappedFields:bookname/book_name mappedProperties:bookName
2、關於underscoreName()的轉換,效果就是: 大寫 –> _+小寫。 初略的認為是轉換成員變量與對應set命名不一樣的問題。
從BeanPropertyRowMapper.initialize(…)結合自己的設想:
1、先根據class緩存了所有的set方法,並保存在了mappedFields。
即當初我想要的效果,不過我想的可能是在執行第一次的時候(mapRow方法中)。而BeanPropertyRowMapper則是在構造的時候就緩存了。
2、我所沒想到的underscoreName(),可能我項目並沒存在命名問題。成員變量的set/get都是工具自己生成的,命名也是采取的駝峰式(不管是java還是sql的別名)
3、對於PropertyDescriptor的獲取,spring還是用了自己的獲取。我的話不確定,畢竟對反射也不是很熟悉。
但看過一篇文章: java反射的性能問題 ,如果我自己寫的話,估計還是會用java自帶的吧。
雖然不清楚spring和java自帶的區別及效率,但我覺得應該spring比較好吧。不然spring直接用jdk的就行了,沒必要自己再寫。
以上,BeanPropertyRowMapper在構造的時候已經有了反射,接下來就是把每行的值寫到對應的屬性中。
二、寫入sql結果集的值到對應bean屬性
// jdbcTemplate調用RowMapper.mapRow(...) public class RowMapperResultSetExtractor<T> implements ResultSetExtractor<List<T>> { private final RowMapper<T> rowMapper; private final int rowsExpected; /** * Create a new RowMapperResultSetExtractor. * @param rowMapper the RowMapper which creates an object for each row */ public RowMapperResultSetExtractor(RowMapper<T> rowMapper) { this(rowMapper, 0); } /** * Create a new RowMapperResultSetExtractor. * @param rowMapper the RowMapper which creates an object for each row * @param rowsExpected the number of expected rows * (just used for optimized collection handling) */ public RowMapperResultSetExtractor(RowMapper<T> rowMapper, int rowsExpected) { Assert.notNull(rowMapper, "RowMapper is required"); this.rowMapper = rowMapper; this.rowsExpected = rowsExpected; } @Override public List<T> extractData(ResultSet rs) throws SQLException { List<T> results = (this.rowsExpected > 0 ? new ArrayList<T>(this.rowsExpected) : new ArrayList<T>()); int rowNum = 0; while (rs.next()) { results.add(this.rowMapper.mapRow(rs, rowNum++));//調用核心; 1、每行的rowMapper是同一個對象,所以可以緩存映射關系 2、mapRow為什么是new對象也是因為這個。不然list.add的是同一個return對象。 } return results; } }
// BeanPropertyRowMapper中mapRow的實現 @Override public T mapRow(ResultSet rs, int rowNumber) throws SQLException { Assert.state(this.mappedClass != null, "Mapped class was not specified"); T mappedObject = BeanUtils.instantiate(this.mappedClass); //實例化一個新對象;就是class.newInstance();
BeanWrapper bw = PropertyAccessorFactory.forBeanPropertyAccess(mappedObject); //這也是spring自己的,有興趣可以看。在這主要就是類似method.invoke(…) initBeanWrapper(bw); //這是個空方法,用於子類擴展
ResultSetMetaData rsmd = rs.getMetaData(); int columnCount = rsmd.getColumnCount(); // 與rsmd都是sql結果集的信息 Set<String> populatedProperties = (isCheckFullyPopulated() ? new HashSet<String>() : null);//是否嚴格映射bean和sql for (int index = 1; index <= columnCount; index++) { String column = JdbcUtils.lookupColumnName(rsmd, index); // 得到sql的列名/別名 PropertyDescriptor pd = this.mappedFields.get(column.replaceAll(" ", "").toLowerCase()); // 從緩存中得到方法信息 if (pd != null) { try { Object value = getColumnValue(rs, index, pd); // 得到每列的值。為什么要pd:因為要根據類型獲取相應的值。 if (logger.isDebugEnabled() && rowNumber == 0) { logger.debug("Mapping column '" + column + "' to property '" + pd.getName() + "' of type " + pd.getPropertyType()); } try { bw.setPropertyValue(pd.getName(), value); // 設置結果 } catch (TypeMismatchException e) { if (value == null && primitivesDefaultedForNullValue) { logger.debug("Intercepted TypeMismatchException for row " + rowNumber + " and column '" + column + "' with value " + value + " when setting property '" + pd.getName() + "' of type " + pd.getPropertyType() + " on object: " + mappedObject); } else { throw e; } } if (populatedProperties != null) { //嚴格映射的邏輯判斷 populatedProperties.add(pd.getName()); } } catch (NotWritablePropertyException ex) { throw new DataRetrievalFailureException( "Unable to map column " + column + " to property " + pd.getName(), ex); } } } if (populatedProperties != null && !populatedProperties.equals(this.mappedProperties)) { //嚴格映射的邏輯判斷 throw new InvalidDataAccessApiUsageException("Given ResultSet does not contain all fields " + "necessary to populate object of class [" + this.mappedClass + "]: " + this.mappedProperties); } return mappedObject; }
public static <T> T instantiate(Class<T> clazz) throws BeanInstantiationException { Assert.notNull(clazz, "Class must not be null"); if (clazz.isInterface()) { throw new BeanInstantiationException(clazz, "Specified class is an interface"); } try { return clazz.newInstance(); } catch (InstantiationException ex) { throw new BeanInstantiationException(clazz, "Is it an abstract class?", ex); } catch (IllegalAccessException ex) { throw new BeanInstantiationException(clazz, "Is the constructor accessible?", ex); } }
/** * Initialize the given BeanWrapper to be used for row mapping. * To be called for each row. * <p>The default implementation is empty. Can be overridden in subclasses. * @param bw the BeanWrapper to initialize */ protected void initBeanWrapper(BeanWrapper bw) { }
protected Object getColumnValue(ResultSet rs, int index, PropertyDescriptor pd) throws SQLException { return JdbcUtils.getResultSetValue(rs, index, pd.getPropertyType()); } // JdbcUtils.class 根據set參數的類型,決定sql返回值調用類型 public static Object getResultSetValue(ResultSet rs, int index, Class<?> requiredType) throws SQLException { if (requiredType == null) { return getResultSetValue(rs, index); } Object value; // Explicitly extract typed value, as far as possible. if (String.class.equals(requiredType)) { return rs.getString(index); } else if (boolean.class.equals(requiredType) || Boolean.class.equals(requiredType)) { value = rs.getBoolean(index); } else if (byte.class.equals(requiredType) || Byte.class.equals(requiredType)) { value = rs.getByte(index); } else if (short.class.equals(requiredType) || Short.class.equals(requiredType)) { value = rs.getShort(index); } else if (int.class.equals(requiredType) || Integer.class.equals(requiredType)) { value = rs.getInt(index); } else if (long.class.equals(requiredType) || Long.class.equals(requiredType)) { value = rs.getLong(index); } else if (float.class.equals(requiredType) || Float.class.equals(requiredType)) { value = rs.getFloat(index); } else if (double.class.equals(requiredType) || Double.class.equals(requiredType) || Number.class.equals(requiredType)) { value = rs.getDouble(index); } else if (BigDecimal.class.equals(requiredType)) { return rs.getBigDecimal(index); } else if (java.sql.Date.class.equals(requiredType)) { return rs.getDate(index); } else if (java.sql.Time.class.equals(requiredType)) { return rs.getTime(index); } else if (java.sql.Timestamp.class.equals(requiredType) || java.util.Date.class.equals(requiredType)) { return rs.getTimestamp(index); } else if (byte[].class.equals(requiredType)) { return rs.getBytes(index); } else if (Blob.class.equals(requiredType)) { return rs.getBlob(index); } else if (Clob.class.equals(requiredType)) { return rs.getClob(index); } else { // Some unknown type desired -> rely on getObject. if (getObjectWithTypeAvailable) { try { return rs.getObject(index, requiredType); } catch (AbstractMethodError err) { logger.debug("JDBC driver does not implement JDBC 4.1 'getObject(int, Class)' method", err); } catch (SQLFeatureNotSupportedException ex) { logger.debug("JDBC driver does not support JDBC 4.1 'getObject(int, Class)' method", ex); } catch (SQLException ex) { logger.debug("JDBC driver has limited support for JDBC 4.1 'getObject(int, Class)' method", ex); } } // Fall back to getObject without type specification... return getResultSetValue(rs, index); } // Perform was-null check if necessary (for results that the JDBC driver returns as primitives). return (rs.wasNull() ? null : value); }
基本核心的全部源碼就在這;但有個核心沒看也沒寫出源碼就是:
BeanWrapper bw = PropertyAccessorFactory.forBeanPropertyAccess(mappedObject);
bw.setPropertyValue(pd.getName(), value);
其目的個人理解就是:method.invoke(…)
感覺總體來說,和我的設想是一樣的。但只是實現上效率的區別。就在PropertyDescriptor、Method、及set調用。
個人話應該不會去寫PropertyDescriptor的獲取,就直接用jdk提供的了。而set的調用也是一樣,直接用PropertyDescriptor找到set的Method,然后Method.invoke(…);
(所以個人最初想的時候緩存的其實是Method而不是PropertyDescriptor)
值的獲取也沒太大區別,都是判斷set的參數類型,然后調用ResultSet的對應方法。
總結:
考慮到了優化最明顯的查找set方法小號。但其余的優化還不知道:不知道哪些還可以優化、怎么優化,有待學習。
(有待不知是哪年了,周圍的都沒注重技術的。都只要求業務功能的代碼,管你怎么寫的,也不願給你時間、資源去學習,就知道催進度。)
ps:
哎,回到家大概19.50。從8.10來分開始邊寫邊看BeanPropertyRowMapper,感覺也沒寫什么而且那源碼核心的復雜的BeanWrapper還沒看。單獨是BeanPropertyRowMapper其實幾分鍾就看完了。結果寫完就22.40左右了。
要說學到了什么嗎,真的不知道。項目實際接觸的都是業務代碼,沒有任何技術可言(不管是舊技術還是新技術)。
心好累…一直想轉行,可惜一無是處,哎!