Registering constants to internal classes (ZE2)

Hello Moriyoshi, to me that sounds good. Tuesday, July 1, 2003, 7:33:08 PM, you wrote: MK> Hi, MK> Perhaps I'm missing something, but I noticed the current ZE2 MK> implementation doesn't seem to correctly handle constants registered to MK> internal classes. I tracked down the cause of this problem, and finally MK> figured out that those constants are destroyed by the ordinary MK> zval_ptr_dtor_wrapper() at the destruction of the constant hash after the MK> deactivation of the memory manager. Then I'm putting forward the following MK> ideas to avoid this. MK> 1. Unify the calls of pefree() / efree() by a function pointer followed by MK> the actual memory block which directs to the actual free() implementation MK> the memory block should be freed by. Here's the example stuff I'm MK> referring to: MK> typedef struct _memblk_t { MK> void (*free_fn)(void *ptr); MK> byte blk[]; MK> } memblk_t; MK> void efree_impl(void *ptr) MK> { MK> ... MK> } MK> void pefree_impl(void *ptr) MK> { MK> free(ptr); MK> } MK> void *pemalloc_impl(size_t sz) MK> { MK> return malloc(sz); MK> } MK> void *pemalloc(size_t sz, int persistent) MK> { MK> void *ptr; MK> if (persistent) { MK> ptr = emalloc_impl(sz + sizeof(memblk_t)); MK> ptr->free_fn = efree_impl; MK> } else { MK> ptr = pemalloc_impl(sz + sizeof(memblk_t)); MK> ptr->free_fn = pefree_impl; MK> } MK> return (void *)ptr->blk; MK> } MK> void efree(void *ptr) MK> { MK> memblk_t *actual_head; MK> actual_head = (memblk_t *)((byte *)ptr - &((memblk_t *)0)->blk); MK> actual_head->free_fn((void *)actual_head); MK> } MK> 2. Add some new constructors / destructors dedicated to the persistent MK> zvals, like ALLOC_INIT_PERSISTENT_ZVAL(), persistent_zval_dtor(), or MK> persistent_zval_ptr_dtor(). I don't like this idea though. MK> Any opinions? MK> Moriyoshi