On 2023-03-07 14:05, Peng Zhang wrote:
Hi, Liam,
- } while (slot > slot_count); + } while (mas->offset >= mas_data_end(mas)); - mas->offset = ++slot; + mt = mte_node_type(mas->node); pivots = ma_pivots(mas_mn(mas), mt); - if (slot > 0) - mas->min = pivots[slot - 1] + 1;
- if (slot <= slot_count) - mas->max = pivots[slot]; + mas->min = pivots[mas->offset] + 1; + mas->offset++; + if (mas->offset < mt_slots[mt]) + mas->max = pivots[mas->offset];
There is a bug here, the assignment of mas->min and mas->max is wrong. The assignment will make them represent the range of a child node, but it should represent the range of the current node. After mas_ascend() returns, mas-min and mas->max already represent the range of the current node, so we should delete these assignments of mas->min and mas->max.
Thanks for your suggestion, Peng. Applying it literally by removing only the min/max assignments:
diff --git a/lib/maple_tree.c b/lib/maple_tree.c index 6fc1ad42b409..9b6e581cf83f 100644 --- a/lib/maple_tree.c +++ b/lib/maple_tree.c @@ -5118,10 +5118,7 @@ static inline bool mas_skip_node
mt = mte_node_type(mas->node); pivots = ma_pivots(mas_mn(mas), mt); - mas->min = pivots[mas->offset] + 1; mas->offset++; - if (mas->offset < mt_slots[mt]) - mas->max = pivots[mas->offset];
return true; }
This allowed my test to pass 100/100 runs. Still in qemu with the test as init, so not really stressed in any way except that specific usecase.
//Snild