Function Prototype

All functions look like this:

void php3_foo(INTERNAL_FUNCTION_PARAMETERS) { }

Even if your function doesn't take any arguments, this is how it is called.

Function Arguments

Arguments are always of type pval. This type contains a union which has the actual type of the argument. So, if your function takes two arguments, you would do something like the following at the top of your function:

Example 48-1. Fetching function arguments pval *arg1, *arg2; if (ARG_COUNT(ht) != 2 || getParameters(ht,2,&arg1,&arg2)==FAILURE) { WRONG_PARAM_COUNT; }

NOTE: Arguments can be passed either by value or by reference. In both cases you will need to pass &(pval *) to getParameters. If you want to check if the n'th parameter was sent to you by reference or not, you can use the function, ParameterPassedByReference(ht,n). It will return either 1 or 0.

When you change any of the passed parameters, whether they are sent by reference or by value, you can either start over with the parameter by calling pval_destructor on it, or if it's an ARRAY you want to add to, you can use functions similar to the ones in internal_functions.h which manipulate return_value as an ARRAY.

Also if you change a parameter to IS_STRING make sure you first assign the new estrdup()'ed string and the string length, and only later change the type to IS_STRING. If you change the string of a parameter which already IS_STRING or IS_ARRAY you should run pval_destructor on it first.

Variable Function Arguments

A function can take a variable number of arguments. If your function can take either 2 or 3 arguments, use the following:

Example 48-2. Variable function arguments pval *arg1, *arg2, *arg3; int arg_count = ARG_COUNT(ht); if (arg_count < 2 || arg_count > 3 || getParameters(ht,arg_count,&arg1,&arg2,&arg3)==FAILURE) { WRONG_PARAM_COUNT; }

Using the Function Arguments

The type of each argument is stored in the pval type field. This type can be any of the following:

If you get an argument of one type and would like to use it as another, or if you just want to force the argument to be of a certain type, you can use one of the following conversion functions:

convert_to_long(arg1); convert_to_double(arg1); convert_to_string(arg1); convert_to_boolean_long(arg1); /* If the string is "" or "0" it becomes 0, 1 otherwise */ convert_string_to_number(arg1); /* Converts string to either LONG or DOUBLE depending on string */

These function all do in-place conversion. They do not return anything.

The actual argument is stored in a union; the members are:

• IS_STRING: arg1->value.str.val

• IS_LONG: arg1->value.lval

• IS_DOUBLE: arg1->value.dval

Memory Management in Functions

Any memory needed by a function should be allocated with either emalloc() or estrdup(). These are memory handling abstraction functions that look and smell like the normal malloc() and strdup() functions. Memory should be freed with efree().

There are two kinds of memory in this program: memory which is returned to the parser in a variable, and memory which you need for temporary storage in your internal function. When you assign a string to a variable which is returned to the parser you need to make sure you first allocate the memory with either emalloc() or estrdup(). This memory should NEVER be freed by you, unless you later in the same function overwrite your original assignment (this kind of programming practice is not good though).

For any temporary/permanent memory you need in your functions/library you should use the three emalloc(), estrdup(), and efree() functions. They behave EXACTLY like their counterpart functions. Anything you emalloc() or estrdup() you have to efree() at some point or another, unless it's supposed to stick around until the end of the program; otherwise, there will be a memory leak. The meaning of "the functions behave exactly like their counterparts" is: if you efree() something which was not emalloc()'ed nor estrdup()'ed you might get a segmentation fault. So please take care and free all of your wasted memory.

If you compile with "-DDEBUG", PHP will print out a list of all memory that was allocated using emalloc() and estrdup() but never freed with efree() when it is done running the specified script.

Setting Variables in the Symbol Table

A number of macros are available which make it easier to set a variable in the symbol table:

• SET_VAR_STRING(name,value)

• SET_VAR_DOUBLE(name,value)

• SET_VAR_LONG(name,value)

Symbol tables in PHP are implemented as hash tables. At any given time, &symbol_table is a pointer to the 'main' symbol table, and active_symbol_table points to the currently active symbol table (these may be identical like in startup, or different, if you're inside a function).

The following examples use 'active_symbol_table'. You should replace it with &symbol_table if you specifically want to work with the 'main' symbol table. Also, the same functions may be applied to arrays, as explained below.

Example 48-3. Checking whether $foo exists in a symbol table if (hash_exists(active_symbol_table,"foo",sizeof("foo"))) { exists... } else { doesn't exist }

Example 48-4. Finding a variable's size in a symbol table hash_find(active_symbol_table,"foo",sizeof("foo"),&pvalue); check(pvalue.type);

Arrays in PHP are implemented using the same hashtables as symbol tables. This means the two above functions can also be used to check variables inside arrays.

If you want to define a new array in a symbol table, you should do the following.

First, you may want to check whether it exists and abort appropriately, using hash_exists() or hash_find().

Next, initialize the array:

Example 48-5. Initializing a new array pval arr; if (array_init(&arr) == FAILURE) { failed... }; hash_update(active_symbol_table,"foo",sizeof("foo"),&arr,sizeof(pval),NULL);

This code declares a new array, named $foo, in the active symbol table. This array is empty.

Here's how to add new entries to it:

Example 48-6. Adding entries to a new array pval entry; entry.type = IS_LONG; entry.value.lval = 5; /* defines $foo["bar"] = 5 */ hash_update(arr.value.ht,"bar",sizeof("bar"),&entry,sizeof(pval),NULL); /* defines $foo[7] = 5 */ hash_index_update(arr.value.ht,7,&entry,sizeof(pval),NULL); /* defines the next free place in $foo[], * $foo[8], to be 5 (works like php2) */ hash_next_index_insert(arr.value.ht,&entry,sizeof(pval),NULL);

If you'd like to modify a value that you inserted to a hash, you must first retrieve it from the hash. To prevent that overhead, you can supply a pval ** to the hash add function, and it'll be updated with the pval * address of the inserted element inside the hash. If that value is NULL (like in all of the above examples) - that parameter is ignored.

hash_next_index_insert() uses more or less the same logic as $foo[] = bar; in PHP 2.0.

If you are building an array to return from a function, you can initialize the array just like above by doing:

if (array_init(return_value) == FAILURE) { failed...; }

...and then adding values with the helper functions:

add_next_index_long(return_value,long_value);
add_next_index_double(return_value,double_value);
add_next_index_string(return_value,estrdup(string_value));

Of course, if the adding isn't done right after the array initialization, you'd probably have to look for the array first:

pval *arr; if (hash_find(active_symbol_table,"foo",sizeof("foo"),(void **)&arr)==FAILURE) { can't find... } else { use arr->value.ht... }

Note that hash_find receives a pointer to a pval pointer, and not a pval pointer.

Just about any hash function returns SUCCESS or FAILURE (except for hash_exists(), which returns a boolean truth value).

Returning simple values

A number of macros are available to make returning values from a function easier.

The RETURN_* macros all set the return value and return from the function:

• RETURN

• RETURN_FALSE

• RETURN_TRUE

• RETURN_LONG(l)

• RETURN_STRING(s,dup) If dup is TRUE, duplicates the string

• RETURN_STRINGL(s,l,dup) Return string (s) specifying length (l).

• RETURN_DOUBLE(d)

The RETVAL_* macros set the return value, but do not return.

• RETVAL_FALSE

• RETVAL_TRUE

• RETVAL_LONG(l)

• RETVAL_STRING(s,dup) If dup is TRUE, duplicates the string

• RETVAL_STRINGL(s,l,dup) Return string (s) specifying length (l).

• RETVAL_DOUBLE(d)

The string macros above will all estrdup() the passed 's' argument, so you can safely free the argument after calling the macro, or alternatively use statically allocated memory.

If your function returns boolean success/error responses, always use RETURN_TRUE and RETURN_FALSE respectively.

Returning complex values

Your function can also return a complex data type such as an object or an array.

Returning an object:

1. Call object_init(return_value).

2. Fill it up with values. The functions available for this purpose are listed below.

3. Possibly, register functions for this object. In order to obtain values from the object, the function would have to fetch "this" from the active_symbol_table. Its type should be IS_OBJECT, and it's basically a regular hash table (i.e., you can use regular hash functions on .value.ht). The actual registration of the function can be done using:

   add_method( return_value, function_name, function_ptr );

The functions used to populate an object are:

• add_property_long( return_value, property_name, l ) - Add a property named 'property_name', of type long, equal to 'l'

• add_property_double( return_value, property_name, d ) - Same, only adds a double

• add_property_string( return_value, property_name, str ) - Same, only adds a string

• add_property_stringl( return_value, property_name, str, l ) - Same, only adds a string of length 'l'

Returning an array:

1. Call array_init(return_value).

2. Fill it up with values. The functions available for this purpose are listed below.

The functions used to populate an array are:

• add_assoc_long(return_value,key,l) - add associative entry with key 'key' and long value 'l'

• add_assoc_double(return_value,key,d)

• add_assoc_string(return_value,key,str,duplicate)

• add_assoc_stringl(return_value,key,str,length,duplicate) specify the string length

• add_index_long(return_value,index,l) - add entry in index 'index' with long value 'l'

• add_index_double(return_value,index,d)

• add_index_string(return_value,index,str)

• add_index_stringl(return_value,index,str,length) - specify the string length

• add_next_index_long(return_value,l) - add an array entry in the next free offset with long value 'l'

• add_next_index_double(return_value,d)

• add_next_index_string(return_value,str)

• add_next_index_stringl(return_value,str,length) - specify the string length

Using the resource list

PHP has a standard way of dealing with various types of resources. This replaces all of the local linked lists in PHP 2.0.

Available functions:

• php3_list_insert(ptr, type) - returns the 'id' of the newly inserted resource

• php3_list_delete(id) - delete the resource with the specified id

• php3_list_find(id,*type) - returns the pointer of the resource with the specified id, updates 'type' to the resource's type

Typical list code would look like this:

Example 48-7. Adding a new resource RESOURCE *resource; /* ...allocate memory for resource and acquire resource... */ /* add a new resource to the list */ return_value->value.lval = php3_list_insert((void *) resource, LE_RESOURCE_TYPE); return_value->type = IS_LONG;

Example 48-8. Using an existing resource pval *resource_id; RESOURCE *resource; int type; convert_to_long(resource_id); resource = php3_list_find(resource_id->value.lval, &type); if (type != LE_RESOURCE_TYPE) { php3_error(E_WARNING,"resource index %d has the wrong type",resource_id->value.lval); RETURN_FALSE; } /* ...use resource... */

Example 48-9. Deleting an existing resource pval *resource_id; RESOURCE *resource; int type; convert_to_long(resource_id); php3_list_delete(resource_id->value.lval);

The resource types should be registered in php3_list.h, in enum list_entry_type. In addition, one should add shutdown code for any new resource type defined, in list.c's list_entry_destructor() (even if you don't have anything to do on shutdown, you must add an empty case).

Using the persistent resource table

PHP has a standard way of storing persistent resources (i.e., resources that are kept in between hits). The first module to use this feature was the MySQL module, and mSQL followed it, so one can get the general impression of how a persistent resource should be used by reading mysql.c. The functions you should look at are:

The general idea of persistence modules is this:

1. Code all of your module to work with the regular resource list mentioned in section (9).

2. Code extra connect functions that check if the resource already exists in the persistent resource list. If it does, register it as in the regular resource list as a pointer to the persistent resource list (because of 1., the rest of the code should work immediately). If it doesn't, then create it, add it to the persistent resource list AND add a pointer to it from the regular resource list, so all of the code would work since it's in the regular resource list, but on the next connect, the resource would be found in the persistent resource list and be used without having to recreate it. You should register these resources with a different type (e.g. LE_MYSQL_LINK for non-persistent link and LE_MYSQL_PLINK for a persistent link).

If you read mysql.c, you'll notice that except for the more complex connect function, nothing in the rest of the module has to be changed.

The very same interface exists for the regular resource list and the persistent resource list, only 'list' is replaced with 'plist':

• php3_plist_insert(ptr, type) - returns the 'id' of the newly inserted resource

• php3_plist_delete(id) - delete the resource with the specified id

• php3_plist_find(id,*type) - returns the pointer of the resource with the specified id, updates 'type' to the resource's type

However, it's more than likely that these functions would prove to be useless for you when trying to implement a persistent module. Typically, one would want to use the fact that the persistent resource list is really a hash table. For instance, in the MySQL/mSQL modules, when there's a pconnect() call (persistent connect), the function builds a string out of the host/user/passwd that were passed to the function, and hashes the SQL link with this string as a key. The next time someone calls a pconnect() with the same host/user/passwd, the same key would be generated, and the function would find the SQL link in the persistent list.

Until further documented, you should look at mysql.c or msql.c to see how one should use the plist's hash table abilities.

One important thing to note: resources going into the persistent resource list must *NOT* be allocated with PHP's memory manager, i.e., they should NOT be created with emalloc(), estrdup(), etc. Rather, one should use the regular malloc(), strdup(), etc. The reason for this is simple - at the end of the request (end of the hit), every memory chunk that was allocated using PHP's memory manager is deleted. Since the persistent list isn't supposed to be erased at the end of a request, one mustn't use PHP's memory manager for allocating resources that go to it.

When you register a resource that's going to be in the persistent list, you should add destructors to it both in the non-persistent list and in the persistent list. The destructor in the non-persistent list destructor shouldn't do anything. The one in the persistent list destructor should properly free any resources obtained by that type (e.g. memory, SQL links, etc). Just like with the non-persistent resources, you *MUST* add destructors for every resource, even it requires no destruction and the destructor would be empty. Remember, since emalloc() and friends aren't to be used in conjunction with the persistent list, you mustn't use efree() here either.

Adding runtime configuration directives

Many of the features of PHP can be configured at runtime. These configuration directives can appear in either the designated php3.ini file, or in the case of the Apache module version in the Apache .conf files. The advantage of having them in the Apache .conf files is that they can be configured on a per-directory basis. This means that one directory may have a certain safemodeexecdir for example, while another directory may have another. This configuration granularity is especially handy when a server supports multiple virtual hosts.

The steps required to add a new directive:

1. Add directive to php3_ini_structure struct in mod_php3.h.

2. In main.c, edit the php3_module_startup function and add the appropriate cfg_get_string() or cfg_get_long() call.

3. Add the directive, restrictions and a comment to the php3_commands structure in mod_php3.c. Note the restrictions part. RSRC_CONF are directives that can only be present in the actual Apache .conf files. Any OR_OPTIONS directives can be present anywhere, include normal .htaccess files.

4. In either php3take1handler() or php3flaghandler() add the appropriate entry for your directive.

5. In the configuration section of the _php3_info() function in functions/info.c you need to add your new directive.

6. And last, you of course have to use your new directive somewhere. It will be addressable as php3_ini.directive.