Delete the currently scanned record

Since the position of the record to be deleted is known, we can just move the last record on the last page to the current slot and decrement current_slotno by 1. If an overflow page is deallocated because of the deletion, a merge is triggered. If the bucket to be merged (the last bucket) is neither the bucket we are scanning nor its image, then the normal merge operation described previously can be used here without any modification. Otherwise, if we still use the same merge operation, the values of current_pageno and current_slotno could be changed due to records movement. In order to minimize the cost and complexity, we designed a special scheme which is illustrated in the following two cases:

• Case 1: the image of the scanned bucket is the last bucket. Since a merge is triggered only when an overflow page is deallocated and we know it's triggered this time after a currently scanned record (in the scanned bucket) is deleted, we can deduce that the scanned bucket consists of some number of full pages. So we can just concatenate all the pages of its image bucket (i.e. the last bucket) to its last page and only the OverflowPageNo on the last page needs to be changed to the PageId of the first page of its image bucket. If its image bucket has only one empty page, nothing needs to be changed. The whole process is illustrated in Figure 2.

   
  Figure: Case 1 in delete_current()
  

• Case 2: the scanned bucket is the last bucket. Again, (as in case 1,) we know that the scanned bucket consists of some number of full pages. We can concatenate all the pages of the image bucket of the scanned bucket to the last page of the scanned bucket as in case 1. But this time the scanned bucket is to be destroyed, Hence we change the PageId of the primary page of the image bucket (on some DirPage) to the PageId of the scanned bucket to keep the whole structure consistent. The whole process is illustrated in Figure 3.

 
Figure: Case 2 in delete_current()

Note that after these operations, all bucket pages are still compact. Besides, the vales of current_pageno and current_slotno are unchanged. One drawback of this scheme is that the continuing scan after the merge will then scan all the pages concatenated to the last page of the scanned bucket where no matched records can be found since they belonged to the image bucket of the scanned bucket before the merge. In order to improve the performance, we set a member of the LinearHashingScan object, last_pageno, to the PageId of the first page concatenated to the scanned bucket. When the scan reads this page, it knows that it's the end of the scan and NOMORERECS is returned. (See Figure 2 and Figure 3.) The above reasoning is based on the fact that when a scan is opened, no update operation is allowed except delete_current().