Re: Const-correctness

Hello Felipe, Sunday, August 3, 2008, 3:12:01 PM, you wrote: > Hi all, > I made a patch that add the const qualifier in several function > parameters in Zend/* where it seems suitable (mostly the API). > http://felipe.ath.cx/diff/const-ness.diff (5_3) > HEAD comming soon in case of no objection. > (And yes... I'm thinking commit it just when HEAD has been done too.) Index: Zend/zend.c =================================================================== -static void print_hash(HashTable *ht, int indent, zend_bool is_object TSRMLS_DC) /* {{{ */ +static void print_hash(HashTable * const ht, int indent, zend_bool is_object TSRMLS_DC) /* {{{ */ -static void print_flat_hash(HashTable *ht TSRMLS_DC) /* {{{ */ +static void print_flat_hash(HashTable * const ht TSRMLS_DC) /* {{{ */ Why not 'const HashTable * const ht', print doesn't sound like it should modify the table itself. -ZEND_API int zend_print_zval(zval *expr, int indent) /* {{{ */ +ZEND_API int zend_print_zval(zval * const expr, int indent) /* {{{ */ -ZEND_API int zend_print_zval_ex(zend_write_func_t write_func, zval *expr, int indent) /* {{{ */ +ZEND_API int zend_print_zval_ex(const zend_write_func_t write_func, zval *expr, int indent) /* {{{ */ -ZEND_API void zend_print_flat_zval_r(zval *expr TSRMLS_DC) /* {{{ */ +ZEND_API void zend_print_flat_zval_r(zval * const expr TSRMLS_DC) /* {{{ */ -ZEND_API void zend_print_zval_r(zval *expr, int indent TSRMLS_DC) /* {{{ */ +ZEND_API void zend_print_zval_r(zval * const expr, int indent TSRMLS_DC) /* {{{ */ -ZEND_API void zend_print_zval_r_ex(zend_write_func_t write_func, zval *expr, int indent TSRMLS_DC) /* {{{ */ +ZEND_API void zend_print_zval_r_ex(const zend_write_func_t write_func, zval *expr, int indent TSRMLS_DC) /* {{{ */ In the same way these should be 'const zval * const expr'. Also note that not all of them use const for expr. Index: Zend/zend_API.c =================================================================== -ZEND_API int zend_copy_parameters_array(int param_count, zval *argument_array TSRMLS_DC) /* {{{ */ +ZEND_API int zend_copy_parameters_array(int param_count, zval * const argument_array TSRMLS_DC) /* {{{ */ 'const zval * const' -ZEND_API int zend_get_object_classname(zval *object, char **class_name, zend_uint *class_name_len TSRMLS_DC) /* {{{ */ +ZEND_API int zend_get_object_classname(zval * const object, char **class_name, zend_uint *class_name_len TSRMLS_DC) /* {{{ */ Does not modify the zval container, so use 'const zval * const object'. -static int parse_arg_object_to_string(zval **arg, char **p, int *pl, int type TSRMLS_DC) /* {{{ */ +static int parse_arg_object_to_string(zval ** const arg, char ** const p, int * const pl, int type TSRMLS_DC) /* {{{ */ This is named parse. Does it convert the zval in place? I rather think it might create a copy of the zval and reduce the refcount in the original one. That would explain wht you cannot do const zval here. So at this point I guess we do that at a bunch of places and sadly we do not have mutable. Maybe we need to do a lot of casting here to make refcount and is_ref work as expected, that is not affect constness of a zval. Maybe some compilers need 'const volatile' for both. If that works you need: 'const zval ** const arg'. -static char *zend_parse_arg_impl(int arg_num, zval **arg, va_list *va, char **spec, char **error, int *severity TSRMLS_DC) /* {{{ */ +static char *zend_parse_arg_impl(int arg_num, zval **arg, va_list * const va, char ** const spec, char **error, int *severity TSRMLS_DC) /* {{{ */ -static int zend_parse_arg(int arg_num, zval **arg, va_list *va, char **spec, int quiet TSRMLS_DC) /* {{{ */ +static int zend_parse_arg(int arg_num, zval **arg, va_list * const va, char ** const spec, int quiet TSRMLS_DC) /* {{{ */ Shouldn't this be 'const char ** const spec'. After all we don't modify the actual string. -static int zend_parse_va_args(int num_args, char *type_spec, va_list *va, int flags TSRMLS_DC) /* {{{ */ +static int zend_parse_va_args(int num_args, char *type_spec, va_list * const va, int flags TSRMLS_DC) /* {{{ */ Shouldn't this be 'const char *type_spec' for the same reason as above? -ZEND_API int zend_parse_parameters_ex(int flags, int num_args TSRMLS_DC, char *type_spec, ...) /* {{{ */ +ZEND_API int zend_parse_parameters_ex(int flags, int num_args TSRMLS_DC, char * const type_spec, ...) /* {{{ */ -ZEND_API int zend_parse_parameters(int num_args TSRMLS_DC, char *type_spec, ...) /* {{{ */ +ZEND_API int zend_parse_parameters(int num_args TSRMLS_DC, char * const type_spec, ...) /* {{{ */ -ZEND_API int zend_parse_method_parameters(int num_args TSRMLS_DC, zval *this_ptr, char *type_spec, ...) /* {{{ */ +ZEND_API int zend_parse_method_parameters(int num_args TSRMLS_DC, zval * const this_ptr, char * const type_spec, ...) /* {{{ */ -ZEND_API int zend_parse_method_parameters_ex(int flags, int num_args TSRMLS_DC, zval *this_ptr, char *type_spec, ...) /* {{{ */ +ZEND_API int zend_parse_method_parameters_ex(int flags, int num_args TSRMLS_DC, zval * const this_ptr, char * const type_spec, ...) /* {{{ */ Shouldn't this be 'const char * const type_spec'. For the method versions it should also be 'const zval * const this_ptr'. -ZEND_API int zend_startup_module_ex(zend_module_entry *module TSRMLS_DC) /* {{{ */ +ZEND_API int zend_startup_module_ex(zend_module_entry * const module TSRMLS_DC) /* {{{ */ Why not 'const zend_module_entry * const module'? -ZEND_API zend_module_entry* zend_register_internal_module(zend_module_entry *module TSRMLS_DC) /* {{{ */ +ZEND_API zend_module_entry* zend_register_internal_module(zend_module_entry * const module TSRMLS_DC) /* {{{ */ If only we had mutable... Overall you do a ton of 'struct * const var' which only means that you are going to copy the pointer explicitly. Now functions that use a pointer in a loop and increment it cannot optimize the code anymore and are forced to use an additional real variable on the stack. So unless you have a very smart compiler, the result is an increased stack size. Generally speaking I prefer const on the reight and mid (between *'s) only. But others prefer it to denote that not even the passed in pointer gets modifed and this sometimes even makes debugging easier. > Since then, anyone have an objection? See above, we might be able to add more const :-) Best regards, Marcus