略微加速

什么是 Zend ? 什么是 PHP ?

Zend是语言引擎，PHP内核。PHP是从外层展现的完整系统。咋一听似乎有点模糊不清，但是其实并不复杂( 看下面).为了实现一个 web 脚本解释器，你需要三个部分：

1. 第一：解释器部分分析输入代码，翻译代码，然后执行代码。

2. 第二：功能部分 完成语言的功能（函数，等等）。

3. 第三：接口部分与web通信，等等。

以下部分将讨论PHP能在哪里扩展并如何扩展。

External Modules

External modules can be loaded at script runtime using the function dl(). This function loads a shared object from disk and makes its functionality available to the script to which it's being bound. After the script is terminated, the external module is discarded from memory. This method has both advantages and disadvantages, as described in the following table:

Third parties might consider using the extension tag in php.ini to create additional external modules to PHP. These external modules are completely detached from the main package, which is a very handy feature in commercial environments. Commercial distributors can simply ship disks or archives containing only their additional modules, without the need to create fixed and solid PHP binaries that don't allow other modules to be bound to them.

Built-in Modules

Built-in modules are compiled directly into PHP and carried around with every PHP process; their functionality is instantly available to every script that's being run. Like external modules, built-in modules have advantages and disadvantages, as described in the following table:

The Zend Engine

Of course, extensions can also be implemented directly in the Zend engine. This strategy is good if you need a change in the language behavior or require special functions to be built directly into the language core. In general, however, modifications to the Zend engine should be avoided. Changes here result in incompatibilities with the rest of the world, and hardly anyone will ever adapt to specially patched Zend engines. Modifications can't be detached from the main PHP sources and are overridden with the next update using the "official" source repositories. Therefore, this method is generally considered bad practice and, due to its rarity, is not covered in this book.

Extension Conventions

Zend is built using certain conventions; to avoid breaking its standards, you should follow the rules described in the following sections.

Macros

For almost every important task, Zend ships predefined macros that are extremely handy. The tables and figures in the following sections describe most of the basic functions, structures, and macros. The macro definitions can be found mainly in zend.h and zend_API.h. We suggest that you take a close look at these files after having studied this chapter. (Although you can go ahead and read them now, not everything will make sense to you yet.)

Memory Management

Resource management is a crucial issue, especially in server software. One of the most valuable resources is memory, and memory management should be handled with extreme care. Memory management has been partially abstracted in Zend, and you should stick to this abstraction for obvious reasons: Due to the abstraction, Zend gets full control over all memory allocations. Zend is able to determine whether a block is in use, automatically freeing unused blocks and blocks with lost references, and thus prevent memory leaks. The functions to be used are described in the following table:

Directory and File Functions

The following directory and file functions should be used in Zend modules. They behave exactly like their C counterparts, but provide virtual working directory support on the thread level.

String Handling

Strings are handled a bit differently by the Zend engine than other values such as integers, Booleans, etc., which don't require additional memory allocation for storing their values. If you want to return a string from a function, introduce a new string variable to the symbol table, or do something similar, you have to make sure that the memory the string will be occupying has previously been allocated, using the aforementioned e*() functions for allocation. (This might not make much sense to you yet; just keep it somewhere in your head for now - we'll get back to it shortly.)

Complex Types

Complex types such as arrays and objects require different treatment. Zend features a single API for these types - they're stored using hash tables.

Note:

To reduce complexity in the following source examples, we're only working with simple types such as integers at first. A discussion about creating more advanced types follows later in this chapter.

Compiling Modules

There are basically two ways to compile modules:

• Use the provided "make" mechanism in the ext directory, which also allows building of dynamic loadable modules.

• Compile the sources manually.

The second method is good for those who (for some reason) don't have the full PHP source tree available, don't have access to all files, or just like to juggle with their keyboard. These cases should be extremely rare, but for the sake of completeness we'll also describe this method.

Compiling Using Make

To compile the sample sources using the standard mechanism, copy all their subdirectories to the ext directory of your PHP source tree. Then run buildconf, which will create an updated configure script containing appropriate options for the new extension. By default, all the sample sources are disabled, so you don't have to fear breaking your build process.

After you run buildconf, configure --help shows the following additional modules:

  --enable-array_experiments   BOOK: Enables array experiments
  --enable-call_userland       BOOK: Enables userland module
  --enable-cross_conversion    BOOK: Enables cross-conversion module
  --enable-first_module        BOOK: Enables first module
  --enable-infoprint           BOOK: Enables infoprint module
  --enable-reference_test      BOOK: Enables reference test module
  --enable-resource_test       BOOK: Enables resource test module
  --enable-variable_creation   BOOK: Enables variable-creation module

The module shown earlier in A simple extension. can be enabled with --enable-first_module or --enable-first_module=yes.

Compiling Manually

To compile your modules manually, you need the following commands:

Note: All include paths in the example are relative to the directory ext. If you're compiling from another directory, change the pathnames accordingly. Required items are the PHP directory, the Zend directory, and (if necessary), the directory in which your module resides.

The link command is also a plain vanilla command instructing linkage as a dynamic module.

You can include optimization options in the compilation command, although these have been omitted in this example (but some are included in the makefile template described in an earlier section).

Note: Compiling and linking manually as a static module into the PHP binary involves very long instructions and thus is not discussed here. (It's not very efficient to type all those commands.)

Module Structure

All PHP modules follow a common structure:

• Header file inclusions (to include all required macros, API definitions, etc.)

• C declaration of exported functions (required to declare the Zend function block)

• Declaration of the Zend function block

• Declaration of the Zend module block

• Implementation of get_module()

• Implementation of all exported functions

Header File Inclusions

The only header file you really have to include for your modules is php.h, located in the PHP directory. This file makes all macros and API definitions required to build new modules available to your code.

Tip: It's good practice to create a separate header file for your module that contains module-specific definitions. This header file should contain all the forward definitions for exported functions and include php.h. If you created your module using ext_skel you already have such a header file prepared.

Declaring Exported Functions

To declare functions that are to be exported (i.e., made available to PHP as new native functions), Zend provides a set of macros. A sample declaration looks like this:

ZEND_FUNCTION ( my_function );

ZEND_FUNCTION declares a new C function that complies with Zend's internal API. This means that the function is of type void and accepts INTERNAL_FUNCTION_PARAMETERS (another macro) as parameters. Additionally, it prefixes the function name with zif. The immediately expanded version of the above definitions would look like this:

void zif_my_function ( INTERNAL_FUNCTION_PARAMETERS );

void zif_my_function( int ht
                    , zval * return_value
                    , zval * this_ptr
                    , int return_value_used
                    , zend_executor_globals * executor_globals
                    );

Since the interpreter and executor core have been separated from the main PHP package, a second API defining macros and function sets has evolved: the Zend API. As the Zend API now handles quite a few of the responsibilities that previously belonged to PHP, a lot of PHP functions have been reduced to macros aliasing to calls into the Zend API. The recommended practice is to use the Zend API wherever possible, as the old API is only preserved for compatibility reasons. For example, the types zval and pval are identical. zval is Zend's definition; pval is PHP's definition (actually, pval is an alias for zval now). As the macro INTERNAL_FUNCTION_PARAMETERS is a Zend macro, the above declaration contains zval. When writing code, you should always use zval to conform to the new Zend API.

The parameter list of this declaration is very important; you should keep these parameters in mind (see Zend's Parameters to Functions Called from PHP for descriptions).

Declaration of the Zend Function Block

Now that you have declared the functions to be exported, you also have to introduce them to Zend. Introducing the list of functions is done by using an array of zend_function_entry. This array consecutively contains all functions that are to be made available externally, with the function's name as it should appear in PHP and its name as defined in the C source. Internally, zend_function_entry is defined as shown in Internal declaration of zend_function_entry..

typedef struct _zend_function_entry {
    char *fname;
    void (*handler)(INTERNAL_FUNCTION_PARAMETERS);
    unsigned char *func_arg_types;
} zend_function_entry;

zend_function_entry firstmod_functions[] =
{
    ZEND_FE(first_module, NULL)
    ZEND_FE_END
};

The macro ZEND_FE (short for 'Zend Function Entry') simply expands to a structure entry in zend_function_entry. Note that these macros introduce a special naming scheme to your functions - your C functions will be prefixed with zif_, meaning that ZEND_FE(first_module) will refer to a C function zif_first_module(). If you want to mix macro usage with hand-coded entries (not a good practice), keep this in mind.

Tip: Compilation errors that refer to functions named zif_*() relate to functions defined with ZEND_FE.

Macros for Defining Functions shows a list of all the macros that you can use to define functions.

Note: You can't use ZEND_FE in conjunction with PHP_FUNCTION, or PHP_FE in conjunction with ZEND_FUNCTION. However, it's perfectly legal to mix ZEND_FE and ZEND_FUNCTION with PHP_FE and PHP_FUNCTION when staying with the same macro set for each function to be declared. But mixing is not recommended; instead, you're advised to use the ZEND_* macros only.

Declaration of the Zend Module Block

This block is stored in the structure zend_module_entry and contains all necessary information to describe the contents of this module to Zend. You can see the internal definition of this module in Internal declaration of zend_module_entry..

typedef struct _zend_module_entry zend_module_entry;
     
    struct _zend_module_entry {
    unsigned short size;
    unsigned int zend_api;
    unsigned char zend_debug;
    unsigned char zts;
    char *name;
    zend_function_entry *functions;
    int (*module_startup_func)(INIT_FUNC_ARGS);
    int (*module_shutdown_func)(SHUTDOWN_FUNC_ARGS);
    int (*request_startup_func)(INIT_FUNC_ARGS);
    int (*request_shutdown_func)(SHUTDOWN_FUNC_ARGS);
    void (*info_func)(ZEND_MODULE_INFO_FUNC_ARGS);
    char *version;

[ Rest of the structure is not interesting here ]

};

In our example, this structure is implemented as follows:

zend_module_entry firstmod_module_entry =
{
    STANDARD_MODULE_HEADER,
    "First Module",
    firstmod_functions,
    NULL, NULL, NULL, NULL, NULL,
    NO_VERSION_YET,
    STANDARD_MODULE_PROPERTIES,
};

For reference purposes, you can find a list of the macros involved in declared startup and shutdown functions in Macros to Declare Startup and Shutdown Functions. These are not used in our basic example yet, but will be demonstrated later on. You should make use of these macros to declare your startup and shutdown functions, as these require special arguments to be passed (INIT_FUNC_ARGS and SHUTDOWN_FUNC_ARGS), which are automatically included into the function declaration when using the predefined macros. If you declare your functions manually and the PHP developers decide that a change in the argument list is necessary, you'll have to change your module sources to remain compatible.

Creation of get_module()

This function is special to all dynamic loadable modules. Take a look at the creation via the ZEND_GET_MODULE macro first:

#if COMPILE_DL_FIRSTMOD
     ZEND_GET_MODULE(firstmod) 
#endif

The function implementation is surrounded by a conditional compilation statement. This is needed since the function get_module() is only required if your module is built as a dynamic extension. By specifying a definition of COMPILE_DL_FIRSTMOD in the compiler command (see above for a discussion of the compilation instructions required to build a dynamic extension), you can instruct your module whether you intend to build it as a dynamic extension or as a built-in module. If you want a built-in module, the implementation of get_module() is simply left out.

get_module() is called by Zend at load time of the module. You can think of it as being invoked by the dl() call in your script. Its purpose is to pass the module information block back to Zend in order to inform the engine about the module contents.

If you don't implement a get_module() function in your dynamic loadable module, Zend will compliment you with an error message when trying to access it.

Implementation of All Exported Functions

Implementing the exported functions is the final step. The example function in first_module looks like this:

ZEND_FUNCTION(first_module)
{
    long parameter;

    if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "l", &parameter) == FAILURE) {
        return;
    }

    RETURN_LONG(parameter);
}

After the declaration, code for checking and retrieving the function's arguments, argument conversion, and return value generation follows (more on this later).

Summary

That's it, basically - there's nothing more to implementing PHP modules. Built-in modules are structured similarly to dynamic modules, so, equipped with the information presented in the previous sections, you'll be able to fight the odds when encountering PHP module source files.

Now, in the following sections, read on about how to make use of PHP's internals to build powerful extensions.

Determining the Number of Arguments

Since PHP doesn't have formal function definitions with support for call syntax checking, and since PHP features variable arguments, sometimes you need to find out with how many arguments your function has been called. You can use the ZEND_NUM_ARGS macro in this case. In previous versions of PHP, this macro retrieved the number of arguments with which the function has been called based on the function's hash table entry, ht, which is passed in the INTERNAL_FUNCTION_PARAMETERS list. As ht itself now contains the number of arguments that have been passed to the function, ZEND_NUM_ARGS has been stripped down to a dummy macro (see its definition in zend_API.h). But it's still good practice to use it, to remain compatible with future changes in the call interface. Note: The old PHP equivalent of this macro is ARG_COUNT.

The following code checks for the correct number of arguments:

if(ZEND_NUM_ARGS() != 2) WRONG_PARAM_COUNT;

"Warning: Wrong parameter count for firstmodule() in /home/www/htdocs/firstmod.php on line 5"

This macro prints a default error message and then returns to the caller. Its definition can also be found in zend_API.h and looks like this:

ZEND_API void wrong_param_count(void);

#define WRONG_PARAM_COUNT { wrong_param_count(); return; }

Retrieving Arguments

Note: New parameter parsing API

This chapter documents the new Zend parameter parsing API introduced by Andrei Zmievski. It was introduced in the development stage between PHP 4.0.6 and 4.1.0.

Parsing parameters is a very common operation and it may get a bit tedious. It would also be nice to have standardized error checking and error messages. Since PHP 4.1.0, there is a way to do just that by using the new parameter parsing API. It greatly simplifies the process of receiving parameters, but it has a drawback in that it can't be used for functions that expect variable number of parameters. But since the vast majority of functions do not fall into those categories, this parsing API is recommended as the new standard way.

The prototype for parameter parsing function looks like this:

int zend_parse_parameters(int num_args TSRMLS_DC, char *type_spec, ...);

zend_parse_parameters() also performs type conversions whenever possible, so that you always receive the data in the format you asked for. Any type of scalar can be converted to another one, but conversions between complex types (arrays, objects, and resources) and scalar types are not allowed.

If the parameters could be obtained successfully and there were no errors during type conversion, the function will return SUCCESS, otherwise it will return FAILURE. The function will output informative error messages, if the number of received parameters does not match the requested number, or if type conversion could not be performed.

Here are some sample error messages:

Here is the full list of type specifiers:

• l - long

• d - double

• s - string (with possible null bytes) and its length

• b - boolean

• r - resource, stored in zval*

• a - array, stored in zval*

• o - object (of any class), stored in zval*

• O - object (of class specified by class entry), stored in zval*

• z - the actual zval*

• | - indicates that the remaining parameters are optional. The storage variables corresponding to these parameters should be initialized to default values by the extension, since they will not be touched by the parsing function if the parameters are not passed.

• / - the parsing function will call SEPARATE_ZVAL_IF_NOT_REF() on the parameter it follows, to provide a copy of the parameter, unless it's a reference.

• ! - the parameter it follows can be of specified type or NULL (only applies to a, o, O, r, and z). If NULL value is passed by the user, the storage pointer will be set to NULL.

The best way to illustrate the usage of this function is through examples:

/* Gets a long, a string and its length, and a zval. */
long l;
char *s;
int s_len;
zval *param;
if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC,
                          "lsz", &l, &s, &s_len, &param) == FAILURE) {
    return;
}

/* Gets an object of class specified by my_ce, and an optional double. */
zval *obj;
double d = 0.5;
if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC,
                          "O|d", &obj, my_ce, &d) == FAILURE) {
    return;
}

/* Gets an object or null, and an array.
   If null is passed for object, obj will be set to NULL. */
zval *obj;
zval *arr;
if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "O!a", &obj, &arr) == FAILURE) {
    return;
}

/* Gets a separated array. */
zval *arr;
if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "a/", &arr) == FAILURE) {
    return;
}

/* Get only the first three parameters (useful for varargs functions). */
zval *z;
zend_bool b;
zval *r;
if (zend_parse_parameters(3, "zbr!", &z, &b, &r) == FAILURE) {
    return;
}

Note that in the last example we pass 3 for the number of received parameters, instead of ZEND_NUM_ARGS(). What this lets us do is receive the least number of parameters if our function expects a variable number of them. Of course, if you want to operate on the rest of the parameters, you will have to use zend_get_parameters_array_ex() to obtain them.

The parsing function has an extended version that allows for an additional flags argument that controls its actions.

int zend_parse_parameters_ex(int flags, int num_args TSRMLS_DC, char *type_spec, ...);

The only flag you can pass currently is ZEND_PARSE_PARAMS_QUIET, which instructs the function to not output any error messages during its operation. This is useful for functions that expect several sets of completely different arguments, but you will have to output your own error messages.

For example, here is how you would get either a set of three longs or a string:

long l1, l2, l3;
char *s;
if (zend_parse_parameters_ex(ZEND_PARSE_PARAMS_QUIET,
                             ZEND_NUM_ARGS() TSRMLS_CC,
                             "lll", &l1, &l2, &l3) == SUCCESS) {
    /* manipulate longs */
} else if (zend_parse_parameters_ex(ZEND_PARSE_PARAMS_QUIET,
                                    ZEND_NUM_ARGS(), "s", &s, &s_len) == SUCCESS) {
    /* manipulate string */
} else {
    php_error(E_WARNING, "%s() takes either three long values or a string as argument",
              get_active_function_name(TSRMLS_C));
    return;
}

With all the abovementioned ways of receiving function parameters you should have a good handle on this process. For even more example, look through the source code for extensions that are shipped with PHP - they illustrate every conceivable situation.

Old way of retrieving arguments (deprecated)

Note: Deprecated parameter parsing API

This API is deprecated and superseded by the new ZEND parameter parsing API.

After having checked the number of arguments, you need to get access to the arguments themselves. This is done with the help of zend_get_parameters_ex():

zval **parameter;

if(zend_get_parameters_ex(1, &parameter) != SUCCESS)
  WRONG_PARAM_COUNT;

zend_get_parameters_ex() accepts at least two arguments. The first argument is the number of arguments to retrieve (which should match the number of arguments with which the function has been called; this is why it's important to check for correct call syntax). The second argument (and all following arguments) are pointers to pointers to pointers to zvals. (Confusing, isn't it?) All these pointers are required because Zend works internally with **zval; to adjust a local **zval in our function,zend_get_parameters_ex() requires a pointer to it.

The return value of zend_get_parameters_ex() can either be SUCCESS or FAILURE, indicating (unsurprisingly) success or failure of the argument processing. A failure is most likely related to an incorrect number of arguments being specified, in which case you should exit with WRONG_PARAM_COUNT.

To retrieve more than one argument, you can use a similar snippet:

zval **param1, **param2, **param3, **param4;
     
if(zend_get_parameters_ex(4, &param1, &param2, &param3, &param4) != SUCCESS)
    WRONG_PARAM_COUNT;

zend_get_parameters_ex() only checks whether you're trying to retrieve too many parameters. If the function is called with five arguments, but you're only retrieving three of them with zend_get_parameters_ex(), you won't get an error but will get the first three parameters instead. Subsequent calls of zend_get_parameters_ex() won't retrieve the remaining arguments, but will get the same arguments again.

Dealing with a Variable Number of Arguments/Optional Parameters

If your function is meant to accept a variable number of arguments, the snippets just described are sometimes suboptimal solutions. You have to create a line calling zend_get_parameters_ex() for every possible number of arguments, which is often unsatisfying.

For this case, you can use the function zend_get_parameters_array_ex(), which accepts the number of parameters to retrieve and an array in which to store them:

zval **parameter_array[4];

/* get the number of arguments */
argument_count = ZEND_NUM_ARGS();

/* see if it satisfies our minimal request (2 arguments) */
/* and our maximal acceptance (4 arguments) */
if(argument_count < 2 || argument_count > 4)
    WRONG_PARAM_COUNT;

/* argument count is correct, now retrieve arguments */
if(zend_get_parameters_array_ex(argument_count, parameter_array) != SUCCESS)
    WRONG_PARAM_COUNT;

A very clever implementation of this can be found in the code handling PHP's fsockopen() located in ext/standard/fsock.c, as shown in PHP's implementation of variable arguments in fsockopen().. Don't worry if you don't know all the functions used in this source yet; we'll get to them shortly.

pval **args[5];
int *sock=emalloc(sizeof(int));
int *sockp;
int arg_count=ARG_COUNT(ht);
int socketd = -1;
unsigned char udp = 0;
struct timeval timeout = { 60, 0 };
unsigned short portno;
unsigned long conv;
char *key = NULL;
FLS_FETCH();

if (arg_count > 5 || arg_count < 2 || zend_get_parameters_array_ex(arg_count,args)==FAILURE) {
    CLOSE_SOCK(1);
    WRONG_PARAM_COUNT;
}

switch(arg_count) {
    case 5:
        convert_to_double_ex(args[4]);
        conv = (unsigned long) (Z_DVAL_PP(args[4]) * 1000000.0);
        timeout.tv_sec = conv / 1000000;
        timeout.tv_usec = conv % 1000000;
        /* fall-through */
    case 4:
        if (!PZVAL_IS_REF(*args[3])) {
            php_error(E_WARNING,"error string argument to fsockopen not passed by reference");
        }
        pval_copy_constructor(*args[3]);
        ZVAL_EMPTY_STRING(*args[3]);
        /* fall-through */
    case 3:
        if (!PZVAL_IS_REF(*args[2])) {
            php_error(E_WARNING,"error argument to fsockopen not passed by reference");
            return;
        }
        ZVAL_LONG(*args[2], 0);
        break;
}

convert_to_string_ex(args[0]);
convert_to_long_ex(args[1]);
portno = (unsigned short) Z_LVAL_P(args[1]);

key = emalloc(Z_STRLEN_P(args[0]) + 10);

fsockopen() accepts two, three, four, or five parameters. After the obligatory variable declarations, the function checks for the correct range of arguments. Then it uses a fall-through mechanism in a switch() statement to deal with all arguments. The switch() statement starts with the maximum number of arguments being passed (five). After that, it automatically processes the case of four arguments being passed, then three, by omitting the otherwise obligatory break keyword in all stages. After having processed the last case, it exits the switch() statement and does the minimal argument processing needed if the function is invoked with only two arguments.

This multiple-stage type of processing, similar to a stairway, allows convenient processing of a variable number of arguments.

Accessing Arguments

To access arguments, it's necessary for each argument to have a clearly defined type. Again, PHP's extremely dynamic nature introduces some quirks. Because PHP never does any kind of type checking, it's possible for a caller to pass any kind of data to your functions, whether you want it or not. If you expect an integer, for example, the caller might pass an array, and vice versa - PHP simply won't notice.

To work around this, you have to use a set of API functions to force a type conversion on every argument that's being passed (see Argument Conversion Functions).

Note: All conversion functions expect a **zval as parameter.

Note:

You can find a demonstration of the behavior in cross_conversion.php on the accompanying CD-ROM.

Using these functions on your arguments will ensure type safety for all data that's passed to you. If the supplied type doesn't match the required type, PHP forces dummy contents on the resulting value (empty strings, arrays, or objects, 0 for numeric values, FALSE for Booleans) to ensure a defined state.

Following is a quote from the sample module discussed previously, which makes use of the conversion functions:

zval **parameter;

if((ZEND_NUM_ARGS() != 1) || (zend_get_parameters_ex(1, &parameter) != SUCCESS))
{
    WRONG_PARAM_COUNT;
}

convert_to_long_ex(parameter);

RETURN_LONG(Z_LVAL_P(parameter));

typedef pval zval;
     
typedef struct _zval_struct zval;

typedef union _zvalue_value {
    long lval;                 /* long value */
    double dval;               /* double value */
    struct {
        char *val;
        int len;
    } str;
    HashTable *ht;             /* hash table value */
    struct {
        zend_class_entry *ce;
        HashTable *properties;
    } obj;
} zvalue_value;

struct _zval_struct {
    /* Variable information */
    zvalue_value value;        /* value */
    unsigned char type;        /* active type */
    unsigned char is_ref;
    short refcount;
};

Actually, pval (defined in php.h) is only an alias of zval (defined in zend.h), which in turn refers to _zval_struct. This is a most interesting structure. _zval_struct is the "master" structure, containing the value structure, type, and reference information. The substructure zvalue_value is a union that contains the variable's contents. Depending on the variable's type, you'll have to access different members of this union. For a description of both structures, see Zend zval Structure, Zend zvalue_value Structure and Zend Variable Type Constants.

To access a long you access zval.value.lval, to access a double you use zval.value.dval, and so on. Because all values are stored in a union, trying to access data with incorrect union members results in meaningless output.

Accessing arrays and objects is a bit more complicated and is discussed later.

Dealing with Arguments Passed by Reference

If your function accepts arguments passed by reference that you intend to modify, you need to take some precautions.

What we didn't say yet is that under the circumstances presented so far, you don't have write access to any zval containers designating function parameters that have been passed to you. Of course, you can change any zval containers that you created within your function, but you mustn't change any zvals that refer to Zend-internal data!

We've only discussed the so-called *_ex() API so far. You may have noticed that the API functions we've used are called zend_get_parameters_ex() instead of zend_get_parameters(), convert_to_long_ex() instead of convert_to_long(), etc. The *_ex() functions form the so-called new "extended" Zend API. They give a minor speed increase over the old API, but as a tradeoff are only meant for providing read-only access.

Because Zend works internally with references, different variables may reference the same value. Write access to a zval container requires this container to contain an isolated value, meaning a value that's not referenced by any other containers. If a zval container were referenced by other containers and you changed the referenced zval, you would automatically change the contents of the other containers referencing this zval (because they'd simply point to the changed value and thus change their own value as well).

zend_get_parameters_ex() doesn't care about this situation, but simply returns a pointer to the desired zval containers, whether they consist of references or not. Its corresponding function in the traditional API, zend_get_parameters(), immediately checks for referenced values. If it finds a reference, it creates a new, isolated zval container; copies the referenced data into this newly allocated space; and then returns a pointer to the new, isolated value.

This action is called zval separation (or pval separation). Because the *_ex() API doesn't perform zval separation, it's considerably faster, while at the same time disabling write access.

To change parameters, however, write access is required. Zend deals with this situation in a special way: Whenever a parameter to a function is passed by reference, it performs automatic zval separation. This means that whenever you're calling a function like this in PHP, Zend will automatically ensure that $parameter is being passed as an isolated value, rendering it to a write-safe state:

my_function(&$parameter);

But this is not the case with regular parameters! All other parameters that are not passed by reference are in a read-only state.

This requires you to make sure that you're really working with a reference - otherwise you might produce unwanted results. To check for a parameter being passed by reference, you can use the macro PZVAL_IS_REF. This macro accepts a zval* to check if it is a reference or not. Examples are given in in Testing for referenced parameter passing..

Example #8 Testing for referenced parameter passing.

zval *parameter;

if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "z", &parameter) == FAILURE)
    return;

/* check for parameter being passed by reference */
if (!PZVAL_IS_REF(parameter)) {
{
    zend_error(E_WARNING, "Parameter wasn't passed by reference");
    RETURN_NULL();
}

/* make changes to the parameter */
ZVAL_LONG(parameter, 10);

Assuring Write Safety for Other Parameters

You might run into a situation in which you need write access to a parameter that's retrieved with zend_get_parameters_ex() but not passed by reference. For this case, you can use the macro SEPARATE_ZVAL, which does a zval separation on the provided container. The newly generated zval is detached from internal data and has only a local scope, meaning that it can be changed or destroyed without implying global changes in the script context:

zval **parameter;
     
/* retrieve parameter */
zend_get_parameters_ex(1, &parameter);

/* at this stage, <parameter> still is connected */
/* to Zend's internal data buffers */

/* make <parameter> write-safe */
SEPARATE_ZVAL(parameter);

/* now we can safely modify <parameter> */
/* without implying global changes */

Note: As you can easily work around the lack of write access in the "traditional" API (with zend_get_parameters() and so on), this API seems to be obsolete, and is not discussed further in this chapter.

Overview

To create new variables that can be seen "from the outside" by the executing script, you need to allocate a new zval container, fill this container with meaningful values, and then introduce it to Zend's internal symbol table. This basic process is common to all variable creations:

zval *new_variable; 

/* allocate and initialize new container */
MAKE_STD_ZVAL(new_variable); 

/* set type and variable contents here, see the following sections */ 

/* introduce this variable by the name "new_variable_name" into the symbol table */
ZEND_SET_SYMBOL(EG(active_symbol_table), "new_variable_name", new_variable); 

/* the variable is now accessible to the script by using $new_variable_name */

The macro MAKE_STD_ZVAL allocates a new zval container using ALLOC_ZVAL and initializes it using INIT_ZVAL. As implemented in Zend at the time of this writing, initializing means setting the reference count to 1 and clearing the is_ref flag, but this process could be extended later - this is why it's a good idea to keep using MAKE_STD_ZVAL instead of only using ALLOC_ZVAL. If you want to optimize for speed (and you don't have to explicitly initialize the zval container here), you can use ALLOC_ZVAL, but this isn't recommended because it doesn't ensure data integrity.

ZEND_SET_SYMBOL takes care of introducing the new variable to Zend's symbol table. This macro checks whether the value already exists in the symbol table and converts the new symbol to a reference if so (with automatic deallocation of the old zval container). This is the preferred method if speed is not a crucial issue and you'd like to keep memory usage low.

Note that ZEND_SET_SYMBOL makes use of the Zend executor globals via the macro EG. By specifying EG(active_symbol_table), you get access to the currently active symbol table, dealing with the active, local scope. The local scope may differ depending on whether the function was invoked from within a function.

If you need to optimize for speed and don't care about optimal memory usage, you can omit the check for an existing variable with the same value and instead force insertion into the symbol table by using zend_hash_update():

zval *new_variable;

/* allocate and initialize new container */
MAKE_STD_ZVAL(new_variable);

/* set type and variable contents here, see the following sections */

/* introduce this variable by the name "new_variable_name" into the symbol table */
zend_hash_update(
    EG(active_symbol_table),
    "new_variable_name",
    strlen("new_variable_name") + 1,
    &new_variable,
    sizeof(zval *),
    NULL
);

The variables generated with the snippet above will always be of local scope, so they reside in the context in which the function has been called. To create new variables in the global scope, use the same method but refer to another symbol table:

zval *new_variable;
     
// allocate and initialize new container
MAKE_STD_ZVAL(new_variable);

//
// set type and variable contents here
//

// introduce this variable by the name "new_variable_name" into the global symbol table
ZEND_SET_SYMBOL(&EG(symbol_table), "new_variable_name", new_variable);

Note: The active_symbol_table variable is a pointer, but symbol_table is not. This is why you have to use EG(active_symbol_table) and &EG(symbol_table) as parameters to ZEND_SET_SYMBOL - it requires a pointer.

Similarly, to get a more efficient version, you can hardcode the symbol table update:

zval *new_variable;

// allocate and initialize new container
MAKE_STD_ZVAL(new_variable);

//
// set type and variable contents here
//

// introduce this variable by the name "new_variable_name" into the global symbol table
zend_hash_update(
    &EG(symbol_table),
    "new_variable_name",
    strlen("new_variable_name") + 1,
    &new_variable,
    sizeof(zval *),
    NULL
);

Note: You can see that the global variable is actually not accessible from within the function. This is because it's not imported into the local scope using global $global_variable; in the PHP source.

Example #9 Creating variables with different scopes.

ZEND_FUNCTION(variable_creation)
{
    zval *new_var1, *new_var2;

    MAKE_STD_ZVAL(new_var1);
    MAKE_STD_ZVAL(new_var2);

    ZVAL_LONG(new_var1, 10);
    ZVAL_LONG(new_var2, 5);

    ZEND_SET_SYMBOL(EG(active_symbol_table), "local_variable", new_var1);
    ZEND_SET_SYMBOL(&EG(symbol_table), "global_variable", new_var2);

    RETURN_NULL();

}

Longs (Integers)

Now let's get to the assignment of data to variables, starting with longs. Longs are PHP's integers and are very simple to store. Looking at the zval.value container structure discussed earlier in this chapter, you can see that the long data type is directly contained in the union, namely in the lval field. The corresponding type value for longs is IS_LONG (see Creation of a long.).

zval *new_long;

MAKE_STD_ZVAL(new_long);

new_long->type = IS_LONG;
new_long->value.lval = 10;

zval *new_long;

MAKE_STD_ZVAL(new_long);
ZVAL_LONG(new_long, 10);

Doubles (Floats)

Doubles are PHP's floats and are as easy to assign as longs, because their value is also contained directly in the union. The member in the zval.value container is dval; the corresponding type is IS_DOUBLE.

zval *new_double;

MAKE_STD_ZVAL(new_double);

new_double->type = IS_DOUBLE;
new_double->value.dval = 3.45;

zval *new_double;

MAKE_STD_ZVAL(new_double);
ZVAL_DOUBLE(new_double, 3.45);

Strings

Strings need slightly more effort. As mentioned earlier, all strings that will be associated with Zend's internal data structures need to be allocated using Zend's own memory-management functions. Referencing of static strings or strings allocated with standard routines is not allowed. To assign strings, you have to access the structure str in the zval.value container. The corresponding type is IS_STRING:

zval *new_string;
char *string_contents = "This is a new string variable";

MAKE_STD_ZVAL(new_string);

new_string->type = IS_STRING;
new_string->value.str.len = strlen(string_contents);
new_string->value.str.val = estrdup(string_contents);

zval *new_string;
char *string_contents = "This is a new string variable";

MAKE_STD_ZVAL(new_string);
ZVAL_STRING(new_string, string_contents, 1);

If you want to truncate the string at a certain position or you already know its length, you can use ZVAL_STRINGL(zval, string, length, duplicate), which accepts an explicit string length to be set for the new string. This macro is faster than ZVAL_STRING and also binary-safe.

To create empty strings, set the string length to 0 and use empty_string as contents:

new_string->type = IS_STRING;
new_string->value.str.len = 0;
new_string->value.str.val = empty_string;

MAKE_STD_ZVAL(new_string);
ZVAL_EMPTY_STRING(new_string);

Booleans

Booleans are created just like longs, but have the type IS_BOOL. Allowed values in lval are 0 and 1:

zval *new_bool;

MAKE_STD_ZVAL(new_bool);

new_bool->type = IS_BOOL;
new_bool->value.lval = 1;

Arrays

Arrays are stored using Zend's internal hash tables, which can be accessed using the zend_hash_*() API. For every array that you want to create, you need a new hash table handle, which will be stored in the ht member of the zval.value container.

There's a whole API solely for the creation of arrays, which is extremely handy. To start a new array, you call array_init().

zval *new_array;

MAKE_STD_ZVAL(new_array);

array_init(new_array);

To add new elements to the array, you can use numerous functions, depending on what you want to do. Zend's API for Associative Arrays, Zend's API for Indexed Arrays, Part 1 and Zend's API for Indexed Arrays, Part 2 describe these functions. All functions return FAILURE on failure and SUCCESS on success.

All these functions provide a handy abstraction to Zend's internal hash API. Of course, you can also use the hash functions directly - for example, if you already have a zval container allocated that you want to insert into an array. This is done using zend_hash_update() for associative arrays (see Adding an element to an associative array.) and zend_hash_index_update() for indexed arrays (see Adding an element to an indexed array.):

zval *new_array, *new_element;
char *key = "element_key";
      
MAKE_STD_ZVAL(new_array);
MAKE_STD_ZVAL(new_element);

array_init(new_array);

ZVAL_LONG(new_element, 10);

if(zend_hash_update(new_array->value.ht, key, strlen(key) + 1, (void *)&new_element, sizeof(zval *), NULL) == FAILURE)
{
    // do error handling here
}

zval *new_array, *new_element;
int key = 2;

MAKE_STD_ZVAL(new_array);
MAKE_STD_ZVAL(new_element);

array_init(new_array);

ZVAL_LONG(new_element, 10);

if(zend_hash_index_update(new_array->value.ht, key, (void *)&new_element, sizeof(zval *), NULL) == FAILURE)
{
    // do error handling here
}

To emulate the functionality of add_next_index_*(), you can use this:

zend_hash_next_index_insert(ht, zval **new_element, sizeof(zval *), NULL)

Note: To return arrays from a function, use array_init() and all following actions on the predefined variable return_value (given as argument to your exported function; see the earlier discussion of the call interface). You do not have to use MAKE_STD_ZVAL on this.

Tip: To avoid having to write new_array->value.ht every time, you can use HASH_OF(new_array), which is also recommended for compatibility and style reasons.

Objects

Since objects can be converted to arrays (and vice versa), you might have already guessed that they have a lot of similarities to arrays in PHP. Objects are maintained with the same hash functions, but there's a different API for creating them.

To initialize an object, you use the function object_init():

zval *new_object;

MAKE_STD_ZVAL(new_object);

if(object_init(new_object) != SUCCESS)
{
    // do error handling here
}

Resources

Resources are a special kind of data type in PHP. The term resources doesn't really refer to any special kind of data, but to an abstraction method for maintaining any kind of information. Resources are kept in a special resource list within Zend. Each entry in the list has a correspondending type definition that denotes the kind of resource to which it refers. Zend then internally manages all references to this resource. Access to a resource is never possible directly - only via a provided API. As soon as all references to a specific resource are lost, a corresponding shutdown function is called.

For example, resources are used to store database links and file descriptors. The de facto standard implementation can be found in the MySQL module, but other modules such as the Oracle module also make use of resources.

Note:

In fact, a resource can be a pointer to anything you need to handle in your functions (e.g. pointer to a structure) and the user only has to pass a single resource variable to your function.

To create a new resource you need to register a resource destruction handler for it. Since you can store any kind of data as a resource, Zend needs to know how to free this resource if its not longer needed. This works by registering your own resource destruction handler to Zend which in turn gets called by Zend whenever your resource can be freed (whether manually or automatically). Registering your resource handler within Zend returns you the resource type handle for that resource. This handle is needed whenever you want to access a resource of this type later and is most of time stored in a global static variable within your extension. There is no need to worry about thread safety here because you only register your resource handler once during module initialization.

The Zend function to register your resource handler is defined as:

ZEND_API int zend_register_list_destructors_ex(rsrc_dtor_func_t ld, rsrc_dtor_func_t pld, char *type_name, int module_number);

There are two different kinds of resource destruction handlers you can pass to this function: a handler for normal resources and a handler for persistent resources. Persistent resources are for example used for database connection. When registering a resource, either of these handlers must be given. For the other handler just pass NULL.

zend_register_list_destructors_ex() accepts the following parameters:

The resource destruction handler (either normal or persistent resources) has the following prototype:

void resource_destruction_handler(zend_rsrc_list_entry *rsrc TSRMLS_DC);

typedef struct _zend_rsrc_list_entry {
     
    void *ptr;
    int type;
    int refcount;

} zend_rsrc_list_entry;

Now we know how to start things, we define our own resource we want register within Zend. It is only a simple structure with two integer members:

typedef struct {
     
    int resource_link;
    int resource_type;

} my_resource;

void my_destruction_handler(zend_rsrc_list_entry *rsrc TSRMLS_DC) {

    // You most likely cast the void pointer to your structure type

    my_resource *my_rsrc = (my_resource *) rsrc->ptr;

    // Now do whatever needs to be done with you resource. Closing
    // Files, Sockets, freeing additional memory, etc.
    // Also, don't forget to actually free the memory for your resource too!

    do_whatever_needs_to_be_done_with_the_resource(my_rsrc);
}

Note:

One important thing to mention: If your resource is a rather complex structure which also contains pointers to memory you allocated during runtime you have to free them before freeing the resource itself!

Now that we have defined

1. what our resource is and

2. our resource destruction handler

1. create a global variable within the extension holding the resource ID so it can be accessed from every function which needs it

2. define the resource name

3. write the resource destruction handler

4. and finally register the handler

// Somewhere in your extension, define the variable for your registered resources.
    // If you wondered what 'le' stands for: it simply means 'list entry'.
    static int le_myresource;

    // It's nice to define your resource name somewhere
    #define le_myresource_name  "My type of resource"

    [...]

    // Now actually define our resource destruction handler
    void my_destruction_handler(zend_rsrc_list_entry *rsrc TSRMLS_DC) {

        my_resource *my_rsrc = (my_resource *) rsrc->ptr;
        do_whatever_needs_to_be_done_with_the_resource(my_rsrc);
    }

    [...]

    PHP_MINIT_FUNCTION(my_extension) {

        // Note that 'module_number' is already provided through the
        // PHP_MINIT_FUNCTION() function definition.

        le_myresource = zend_register_list_destructors_ex(my_destruction_handler, NULL, le_myresource_name, module_number);

        // You can register additional resources, initialize
        // your global vars, constants, whatever.
    }

To actually register a new resource you use can either use the zend_register_resource() function or the ZEND_REGISTER_RESOURE() macro, both defined in zend_list.h. Although the arguments for both map 1:1 it's a good idea to always use macros to be upwards compatible:

int ZEND_REGISTER_RESOURCE(zval *rsrc_result, void *rsrc_pointer, int rsrc_type);

What is really going on when you register a new resource is it gets inserted in an internal list in Zend and the result is just stored in the given zval * container:

rsrc_id = zend_list_insert(rsrc_pointer, rsrc_type);
     
    if (rsrc_result) {
        rsrc_result->value.lval = rsrc_id;
        rsrc_result->type = IS_RESOURCE;
    }

    return rsrc_id;

    RETURN_RESOURCE(rsrc_id)

Note:

It is common practice that if you want to return the resource immediately to the user you specify the return_value as the zval * container.

Zend now keeps track of all references to this resource. As soon as all references to the resource are lost, the destructor that you previously registered for this resource is called. The nice thing about this setup is that you don't have to worry about memory leakages introduced by allocations in your module - just register all memory allocations that your calling script will refer to as resources. As soon as the script decides it doesn't need them anymore, Zend will find out and tell you.

Now that the user got his resource, at some point he is passing it back to one of your functions. The value.lval inside the zval * container contains the key to your resource and thus can be used to fetch the resource with the following macro: ZEND_FETCH_RESOURCE:

ZEND_FETCH_RESOURCE(rsrc, rsrc_type, rsrc_id, default_rsrc_id, resource_type_name, resource_type)

To force removal of a resource from the list, use the function zend_list_delete(). You can also force the reference count to increase if you know that you're creating another reference for a previously allocated value (for example, if you're automatically reusing a default database link). For this case, use the function zend_list_addref(). To search for previously allocated resource entries, use zend_list_find(). The complete API can be found in zend_list.h.

Macros for Automatic Global Variable Creation

In addition to the macros discussed earlier, a few macros allow easy creation of simple global variables. These are nice to know in case you want to introduce global flags, for example. This is somewhat bad practice, but Table Macros for Global Variable Creation describes macros that do exactly this task. They don't need any zval allocation; you simply have to supply a variable name and value.

Creating Constants

Zend supports the creation of true constants (as opposed to regular variables). Constants are accessed without the typical dollar sign ($) prefix and are available in all scopes. Examples include TRUE and FALSE, to name just two.

To create your own constants, you can use the macros in Macros for Creating Constants. All the macros create a constant with the specified name and value.

You can also specify flags for each constant:

• CONST_CS - This constant's name is to be treated as case sensitive.

• CONST_PERSISTENT - This constant is persistent and won't be "forgotten" when the current process carrying this constant shuts down.

 // register a new constant of type "long"
     REGISTER_LONG_CONSTANT("NEW_MEANINGFUL_CONSTANT", 324, CONST_CS |
     CONST_PERSISTENT); 

zend_printf()

zend_printf() works like the standard printf(), except that it prints to Zend's output stream.

zend_error()

zend_error() can be used to generate error messages. This function accepts two arguments; the first is the error type (see zend_errors.h), and the second is the error message.

zend_error(E_WARNING, "This function has been called with empty arguments");

Including Output in phpinfo()

After creating a real module, you'll want to show information about the module in phpinfo() (in addition to the module name, which appears in the module list by default). PHP allows you to create your own section in the phpinfo() output with the ZEND_MINFO() function. This function should be placed in the module descriptor block (discussed earlier) and is always called whenever a script calls phpinfo().

PHP automatically prints a section in phpinfo() for you if you specify the ZEND_MINFO function, including the module name in the heading. Everything else must be formatted and printed by you.

Typically, you can print an HTML table header using php_info_print_table_start() and then use the standard functions php_info_print_table_header() and php_info_print_table_row(). As arguments, both take the number of columns (as integers) and the column contents (as strings). Source code and screenshot for output in phpinfo. shows a source example and its output. To print the table footer, use php_info_print_table_end().

Example #13 Source code and screenshot for output in phpinfo().

php_info_print_table_start();
php_info_print_table_header(2, "First column", "Second column");
php_info_print_table_row(2, "Entry in first row", "Another entry");
php_info_print_table_row(2, "Just to fill", "another row here");
php_info_print_table_end();

Execution Information

You can also print execution information, such as the current file being executed. The name of the function currently being executed can be retrieved using the function get_active_function_name(). This function returns a pointer to the function name and doesn't accept any arguments. To retrieve the name of the file currently being executed, use zend_get_executed_filename(). This function accesses the executor globals, which are passed to it using the TSRMLS_C macro. The executor globals are automatically available to every function that's called directly by Zend (they're part of the INTERNAL_FUNCTION_PARAMETERS described earlier in this chapter). If you want to access the executor globals in another function that doesn't have them available automatically, call the macro TSRMLS_FETCH() once in that function; this will introduce them to your local scope.

Finally, the line number currently being executed can be retrieved using the function zend_get_executed_lineno(). This function also requires the executor globals as arguments. For examples of these functions, see Printing execution information..

Example #14 Printing execution information.

zend_printf("The name of the current function is %s<br>", get_active_function_name(TSRMLS_C));
zend_printf("The file currently executed is %s<br>", zend_get_executed_filename(TSRMLS_C));
zend_printf("The current line being executed is %i<br>", zend_get_executed_lineno(TSRMLS_C));

config.m4

The file config.m4 is processed by buildconf and must contain all the instructions to be executed during configuration. For example, these can include tests for required external files, such as header files, libraries, and so on. PHP defines a set of macros that can be used in this process, the most useful of which are described in M4 Macros for config.m4.