From: anton.txt@g{oogle}mail.com   
      
   Erland Sommarskog to Anton Shepelev:   
      
   >>Thank  you for the replies, Erland and rja. I have   
   >>found what I think is an ambiguity in the documen-   
   >>ta- tion:   
   >>   
   >>   Locks  placed on a resource are associated with   
   >>   either the current transaction or  the  current   
   >>   session.   
   >>   
   >>where  the  notion  of  "current  transaction"  is   
   >>rather unintuitive, for it refers not to the scope   
   >>of  the  immediate enfolding BEGIN TRAN block, but   
   >>to the outermost one.   
   >   
   >There isn't really any ambiguity, because there  is   
   >only  one transaction.  The inner BEGIN TRANSACTION   
   >does not start any new transaction,  but  only  in-   
   >creases @@trancount.  In the same manner, the first   
   >COMMIT in your example only decreases  @@trancount,   
   >but nothing is actually committed.   
      
   Indeed.   An  inner  BEGIN TRAN  only increments the   
   counter, an inner COMMIT TRAN decrements it, but (!)   
   ROLLBACK TRAN  does  actually  undo  the  the  outer   
   transaction and sets the counter to zero.   This  is   
   all  very  unintuitive  to  me, and requires special   
   handling of ROLLBACKs in both  T-SQL  code  and  the   
   client application.   
      
   --   
   ()  ascii ribbon campaign - against html e-mail   
   /\  http://preview.tinyurl.com/qcy6mjc [archived]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)