Re: [PATCH v2] rbtree: fix the red root

[Qian Cai]

On 1/11/19 6:16 PM, Matthew Wilcox wrote:
> On Fri, Jan 11, 2019 at 03:58:43PM -0500, Qian Cai wrote:
>> diff --git a/lib/rbtree_test.c b/lib/rbtree_test.c
>> index b7055b2a07d3..afad0213a117 100644
>> --- a/lib/rbtree_test.c
>> +++ b/lib/rbtree_test.c
>> @@ -345,6 +345,17 @@ static int __init rbtree_test_init(void)
>>  		check(0);
>>  	}
>>  
>> +	/*
>> +	 * a little regression test to catch a bug may be introduced by
>> +	 * 6d58452dc06 (rbtree: adjust root color in rb_insert_color() only when
>> +	 * necessary)
>> +	 */
>> +	insert(nodes, &root);
>> +	nodes->rb.__rb_parent_color = RB_RED;
>> +	insert(nodes + 1, &root);
>> +	erase(nodes + 1, &root);
>> +	erase(nodes, &root);
> 
> That's not a fair test!  You're poking around in the data structure to
> create the situation.  This test would have failed before 6d58452dc06 too.
> How do we create a tree that has a red parent at root, only using insert()
> and erase()?
> 

If only I knew how to reproduce this myself, I might be able to figure out how
it ends up with the red root in the first place.