元数据：跨引擎超完备字段级血缘关系解题方法

INSERT OVERWRITE TABLE DM.DM_CHANNEL_MAIN_1D_DI PARTITION(SDT)
SELECT
      USER_ID
FROM ROWNUMBER

//单独添加Insert into语句
insertIO
    : INSERT (INTO|OVERWRITE) TABLE qualifiedName partitionSpec? query
    ;
//分区表达式以及分区内的表达式
partitionSpec
    : PARTITION '(' partitionVal (',' partitionVal)* ')'
    ;
partitionVal
    : identifier (EQ identifier)?
    ;

SELECT
TRIM(CONCAT('001'
        ,IF(CAST (GROUPING__ID AS INT) || FLOOR(POWER(2,13)) = 0 ,'1','')
        ,IF(CAST (GROUPING__ID AS INT) & FLOOR(POWER(2,11)) = 0 ,'2','')
 FROM TABLE1 

valueExpression
    : primaryExpression                                                                 #valueExpressionDefault
    | valueExpression AT timeZoneSpecifier                                              #atTimeZone
    | operator=(MINUS | PLUS) valueExpression                                           #arithmeticUnary
    | left=valueExpression operator=(ASTERISK | SLASH | PERCEN) right=valueExpression  #arithmeticBinary
    | left=valueExpression operator=(PLUS | MINUS) right=valueExpression                #arithmeticBinary
    | left=valueExpression CONCAT right=valueExpression                                 #concatenation
    ;
CONCAT: '||' | '&';//这里添加’&‘并写法

INSERT OVERWRITE TABLE DM.DM_TRAFFIC_CHANNEL_MAIN_1D_DI
SELECT
T1.AA,T1.CC,T2.FF
FROM (
SELECT
TA.AA+TA.BB AS AA,
CASE WHEN TB.CC='A' THEN TA.CC ELSE TB.AA END + CASE WHEN TB.CC='A' THEN TA.CC ELSE TB.AA END AS CC,
SUBSTRING(TB.CC,1,2) AS EEE,
SUBSTRING(HH,1,2) AS FF
FROM DIM.ROWNUMBER_INFO AS TA
LEFT JOIN DWS.CONSUMER_INFO AS TB  ON TA.A=TB.B
) T1
LEFT JOIN TEST_INFO T2 ON T1.A = T2.A

public class Field {
    public String fieldID;//有字段别名就取字段别名，没别名就用字段名称作为fieldID
    public String fieldName;
    public String tableName;
    public String tableAlias;
    public String dataBase;
    public String isSubQuery="0";//"1" 实体表；"0" 子查询
}

public class LogicField {
    public String logicFiedlID;//存放字段别名，没有字段别名的字段名
    public List<Field> fieldObjList;//拆解多个字段，并把数据库、表等信息完善后，把Field对象列表回写
    public String fieldContent;//存放此字段的实际内容
    public List<String> fieldList = new ArrayList<>(); //实际拆解出字段列表
    public List<String> tableList = new ArrayList<>();//实际拆解出表列表，在遍历LogicField字段会判断是否为子查询，用真实数据库来覆盖
    public List<String> tableAliasList = new ArrayList<>();//实际拆解出表-别名称列表，用于匹配上下层血缘字段构成唯一性查找的问题
    public List<String>  dataBaseList  = new ArrayList<>() ;//作用同tableList；
    public String isMultiFieldFlag="0"; //是否是多个字段组成 "1" 代表 是；"0"代表 否
}

 这里设置三个变量
     1）UnionCnt UNion的个数 在SetOperation函数中去判断
     2）isUnionFlag中判断是否有Uinon操作
     3）UnionStep步长判断深度优先遍历的Union扫描何时结束，把此次的开关关上，计数器复位为0。

if(isUnionFlag && (UnionStep < UnionCnt)){
            if((InvViable != 0)&&(UnionStep == 0)){//如果这个Union不是在最内层0开始的，说明前面已经有层级存在，需要+1 上升一个层级，但是union内部操作多次不变化,第一次加过不再加了
                InvViable = InvViable + 1;
            }else{
                isStartFlag = false;//启动之后，就不再使用0的层级，之后每个层级都自动+1
            }
            if(querySpecificationLevel.keySet().contains(InvViable)){//如果这里Union是在同一调用层级，需要里面的LogicField集合拿出来进行合并操作
                Multimap<String,LogicField> logicFieldMapOther = querySpecificationLevel.get(InvViable);//相同别名ID，可能对应多个LogicField对象
                logicFieldMap.putAll(logicFieldMapOther);
                querySpecificationLevel.put(InvViable,logicFieldMap);
            }else{
                querySpecificationLevel.put(InvViable,logicFieldMap);
            }
            UnionStep++;
        }else{
            UnionStep = 0;
            UnionCnt = 0;
            isUnionFlag = false;
            if(InvViable == 0 && isStartFlag){//如果是从非0开始的 累计+1
                querySpecificationLevel.put(InvViable,logicFieldMap);
                isStartFlag = false;//启动之后，就不再使用0的层级，之后每个层级都自动+1
            }else {
                InvViable = InvViable + 1;
                querySpecificationLevel.put(InvViable,logicFieldMap);
            }

WITH BODY AS ( 
SELECT 
  TA.AA,
  TA.BB,
  TB.CC,
  TB.DD,
  TB.EE      
FROM DIM.ROWNUMBER_INFO AS TA  
LEFT JOIN DWS.CONSUMER_INFO AS TB  
ON TA.A=TB.B AND TA.C='1' AND TB.D = 'AAAAA' ) 
INSERT OVERWRITE TABLE DM.DM_TRAFFIC_CHANNEL_MAIN_1D_DI PARTITION(SDT) 
SELECT  AA,BB,CC,DD,EE FROM BODY AS W

    /*
     *  把with  别名称  as 子查询（）改写成：(表别名,子查询) 键值对。后面有引用别名的，直接替换为子查询。
     * @param ctx
     */
    @Override
    public String visitWith(HiveSqlBaseParser.WithContext ctx) {
        StringBuffer sb = null;
        List<HiveSqlBaseParser.NamedQueryContext> namedQueryContexts = ctx.namedQuery();
        for(HiveSqlBaseParser.NamedQueryContext context : namedQueryContexts){
            sb = new StringBuffer();
            for(int i=1;i<context.getChildCount();i++){
                if(!context.getChild(i).getText().equals("AS")){
                    sb.append(" " + visitResult(context.getChild(i)) );
                }
            }
            sourceTableMapWiht.put(context.getChild(0).getText(),"(" +sb.toString() + ")");
        }
        return  ""; //这里就把with 别名 as （）子查询的写法直接删掉，
    }

    public String visitTableName(HiveSqlBaseParser.TableNameContext ctx) {
        String curTableName = ctx.qualifiedName().getText();
        String replacedSubQueryTableName = null;
        if(sourceTableMapWiht.containsKey(curTableName)){
            replacedSubQueryTableName = sourceTableMapWiht.get(curTableName);
        }else{
            replacedSubQueryTableName = curTableName;
        }
        return replacedSubQueryTableName;
    }