Oracle的递归汇总让我头疼

初衷是做一个关于行政区划的递归汇总，但加上按阶段类别分类汇总的条件就感到很棘手，目前在iBatis里只好这么写

<select id="getRecordsByCondition" parameterClass ="java.util.Map" resultMap="Stats-result">
 <dynamic>	
	<isEqual property="LEVEL" compareValue="1">
		<![CDATA[
		select * from (
	select t.xzqhdm,'该地直辖' as xzqh,t.hs,t.fwmj,t.lgmsl,t.fssssl,t.rk,t.tdmj,t.jdlb
	 from YM_MV_NCYMCC t
	 ]]>
	 <isNotEmpty prepend=' where ' property='PXZQHDM'>               
	     	<![CDATA[t.xzqhdm=#PXZQHDM:VARCHAR#]]>
		</isNotEmpty>
		<isNotEmpty prepend=' and ' property='JDLB'>               
	     	<![CDATA[t.jdlb=#JDLB:CHAR#]]>
		</isNotEmpty>
		<![CDATA[
  		union all
		select t.xzqhdm,t.xzqh,	
       (select sum(hs)
         from YM_MV_NCYMCC t2
         start with t.xzqhdm=t2.xzqhdm and t2.jdlb=#JDLB#
         connect by prior xzqhdm=pxzqhdm and t2.jdlb=#JDLB#
         ) hs,
         (select sum(fwmj)
         from YM_MV_NCYMCC t2
         start with t.xzqhdm=t2.xzqhdm and t2.jdlb=#JDLB#
         connect by prior xzqhdm=pxzqhdm and t2.jdlb=#JDLB#
         ) fwmj,
         (select sum(lgmsl)
         from YM_MV_NCYMCC t2
         start with t.xzqhdm=t2.xzqhdm and t2.jdlb=#JDLB#
         connect by prior xzqhdm=pxzqhdm and t2.jdlb=#JDLB#
         ) lgmsl,
         (select sum(fssssl)
         from YM_MV_NCYMCC t2
         start with t.xzqhdm=t2.xzqhdm and t2.jdlb=#JDLB#
         connect by prior xzqhdm=pxzqhdm and t2.jdlb=#JDLB#
         ) fssssl,
         (select sum(rk)
         from YM_MV_NCYMCC t2
         start with t.xzqhdm=t2.xzqhdm and t2.jdlb=#JDLB#
         connect by prior xzqhdm=pxzqhdm and t2.jdlb=#JDLB#
         ) rk,     
         (select sum(tdmj)
         from YM_MV_NCYMCC t2
         start with t.xzqhdm=t2.xzqhdm and t2.jdlb=#JDLB#
         connect by prior xzqhdm=pxzqhdm and t2.jdlb=#JDLB#
         ) tdmj,
         t.jdlb 
	from YM_MV_NCYMCC t
		]]>
		<isNotEmpty prepend=' where ' property='XZQHDM'>               
	     	<![CDATA[t.XZQHDM like #XZQHDM:VARCHAR#]]>
		</isNotEmpty>
		<isNotEmpty prepend='and' property='PXZQHDM'>               
	     	<![CDATA[t.XZQHDM<>#PXZQHDM:VARCHAR#]]>
		</isNotEmpty>
		<isNotEmpty prepend=' and ' property='JDLB'>               
	     	<![CDATA[t.jdlb=#JDLB:CHAR#]]>
		</isNotEmpty>
		<![CDATA[
		) v
		]]>		
		<isNotEmpty prepend='and' property='QUERYVAL'>               
	     	<![CDATA[v.$FIELD$$operator$#QUERYVAL:NUMBER#]]>
		</isNotEmpty>		
	</isEqual>
 </dynamic>   
</select>

这个同样的递归执行了几遍，逻辑应该可以优化下，可我脑袋已经很难再转动了...
做点小结:
Select * from …. Where [结果过滤条件语句]
Start with  [and起始条件过滤语句]
Connect by prior [and中间记录过滤条件语句]
connect by prior xzqhdm=pxzqhdm 采用的是从根往叶汇总的方式