From: Jorgen Loland
Date: March 16 2011 2:45pm
Subject: Re: bzr commit into mysql-trunk branch (jorgen.loland:3279)
List-Archive: http://lists.mysql.com/commits/133133
Message-Id: <4D80CD07.7010405@oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

>> In the testcase (added to the .inc file), TRP_RANGE::key is set here 
>> (opt_range.cc:5367):
>>
>>   if (key_to_read)
>>   {
>>     idx= key_to_read - tree->keys;
>>     if ((read_plan= new (param->mem_root) TRP_RANGE(*key_to_read, idx,
>>                                                     best_mrr_flags)))
>> (the constructor sets TRP_RANGE::key to *key_to_read).
>> Is there, in the existing code, some place where we take this 
>> TRP_RANGE::key and go up following the "prev" pointers?
> 
> When constructing QUICK_SELECT (get_quick_keys()) we recursively call 
> get_quick_keys() on the left-pointer until it's left pointer is 
> null_element. So this is a good point... maybe I should replace prev 
> with left (and then do the !null_element thingy) in my patch. Do you 
> prefer that?

Thinking a bit more about it, there will be fewer steps to reach the 
first range if I walk down the left pointer instead of prev. Probably 
not worth much considering that we normally don't have tens of ranges to 
walk down, but log2(N) is better than N/2 even when N is small...

So I will change it...