The reason is that each thread can obtain its own distinct and therefore safe allocator.
The model is such that when the thread for finishes a unit of work, it returns the memory to the thread's allocator for the next unit of work. The alternative is the malloc/free model which you are used to, which has performance penalties for allocating from a common and therefore contested/mutexed resource. Hope this explanation is helpful. We can and should do a better job of documenting such, Nick may have a chapter in his APR book on the subject. On Jan 15, 2017 14:16, "Sergey Safarov" <s.safa...@gmail.com> wrote: Could you help me understand why not used mutex to implement thread safety functions: 1. apr_palloc 2. apr_pool_clear 3. apr_pool_destroy 4. apr_pvsprintf In this functions is used: 1. pool_concurrency_set_used 2. pool_concurrency_set_idle 3. pool_concurrency_set_destroyed And if concurrency access is detected then called "pool_concurrency_abort" and program terminated. https://github.com/apache/apr/blob/trunk/memory/unix/apr_pools.c
