php-src/UPGRADING.INTERNALS

PHP 7.4 INTERNALS UPGRADE NOTES

1. Internal API changes
  a. php_sys_symlink() and php_sys_link()
  b. zend_lookup_class_ex() and zend_fetch_class_by_name()
  c. Function/property/class flags
  d. Removed zend_check_private()
  e. php_win32_error_to_msg() memory management
  f. get_properties_for() handler / Z_OBJDEBUG_P
  g. Required object handlers
  h. Immutable classes and op_arrays
  i. php_fgetcsv() and php_fputcsv()
  j. Removed add_get_assoc_*() and add_get_index_*()
  k. Class declaration opcodes
  l. HASH_FLAG_INITIALIZED
  m. write_property return value
  n. Assignments to references
  o. ZEND_COMPILE_EXTENDED_INFO split
  p. ZEND_EXT_FCALL_BEGIN can access arguments
  q. ZEND_COMPILE_IGNORE_USER_FUNCTIONS and ZEND_COMPILE_IGNORE_INTERNAL_FUNCTIONS
  r. TSRM environment locking
  s. Typed references support
  t. Exceptions thrown by string conversions.
  u. Removed uint and ulong typedefs
  v. Compound assignment opcodes
  w. APACHE symbol removed
  x. php_error_docref0() renamed to php_error_docref()
  y. Stream wrapper read/write ops

2. Build system changes
  a. Abstract
  b. Unix build system changes
  c. Windows build system changes

3. Module changes
  a. ext/xml
  b. ext/hash

========================
1. Internal API changes
========================

 a. php_sys_symlink() and php_sys_link() portability macros have been
    added, which behave like POSIX's symlink() and link(), respectively, on
    POSIX compliant systems and on Windows.

 b. zend_lookup_class_ex() and zend_fetch_class_by_name() prototypes were
    changed to accept optional lower-case class name as zend_string*,
    instead of zval*.

 c. Function/property/class flags changes
    - ZEND_ACC_IMPLEMENTED_ABSTRACT is removed (it was used only internally
      during inheritance).
    - ZEND_ACC_IMPLICIT_PUBLIC is removed (it was used only for reflection)
    - ZEND_ACC_SHADOW property flag is removed. Instead of creating shadow
      clone, now we use the same private property_info, and should also
      check property_info->ce (in the same way as with methods).
    - ZEND_ACC_ANON_BOUND is replaced with ZEND_ACC_LINKED. This flag is set
      not only during anonymous classes declaration, but also during any
      run-time or compile-time class declaration.
    - ZEND_ACC_NO_RT_ARENA renamed into ZEND_ACC_HEAP_RT_CACHE. Now it's used
      not only for closures, but also for pseudo-main op_arrays.
    - ZEND_ACC_... flags are re-numbered.

  d. zend_check_private() is removed. Use (func->common.scope == scope) instead.

  e. Pointers returned by php_win32_error_to_msg() have to be freed using
     php_win32_error_msg_free(). Same regarding php_win_err() vs.
     php_win_err_free().

  f. A new, optional object handler with the signature

         HashTable *get_properties_for(zval *obj, zend_prop_purpose purpose)

     has been introduced, where zend_prop_purpose (currently) takes one of:

         ZEND_PROP_PURPOSE_DEBUG       // var_dump etc.
         ZEND_PROP_PURPOSE_ARRAY_CAST  // (array) $obj
         ZEND_PROP_PURPOSE_SERIALIZE   // "O"-format serialization (__wakeup)
         ZEND_PROP_PURPOSE_VAR_EXPORT  // var_export (__set_state)
         ZEND_PROP_PURPOSE_JSON        // json_encode

     The handler returns a non-null HashTable with increased refcounted, and
     the return value must be released using zend_release_properties().

     This handler serves the same general function as get_properties(), but
     provides more control over different property uses, while also making
     it possible to return a temporary property table.

     get_properties() is still used in cases where none of the above purposes
     apply, but overloading get_properties() is generally discouraged. If you
     want to provide purposes for general usage rather than just debugging or
     serialization, please prefer using properly declared properties.

     get_debug_info() is superseded by get_properties_for() with the
     ZEND_PROP_PURPOSE_DEBUG purpose, but remains available for backwards-
     compatibility reasons. However, while it is fine to define this handler,
     it should never be directly called by consuming code.

     The Z_OBJDEBUG_P macro has been removed. It should be replaced by calls to
     zend_get_properties_for() with the ZEND_PROP_PURPOSE_DEBUG purpose:

         // OLD
         int is_temp;
         HashTable *ht = Z_OBJDEBUG_P(obj, is_temp);
         // ...
         if (is_temp) {
            zend_hash_destroy(ht);
            FREE_HASHTABLE(ht);
         }

         // NEW
         HashTable *ht = zend_get_properties_for(obj, ZEND_PROP_PURPOSE_DEBUG);
         // ...
         zend_release_properties(ht);

  g. The following object handlers are now required (must be non-NULL):

       * free_obj
       * dtor_obj
       * read_property
       * write_property
       * read_dimension
       * write_dimension
       * get_property_ptr_ptr
       * has_property
       * unset_property
       * has_dimension
       * unset_dimension
       * get_properties
       * get_method
       * get_constructor
       * get_class_name
       * get_gc

     It is recommended to initialize object handler structures by copying the
     std object handlers and only overwriting those you want to change.

  h. Opcache may make classes and op_arrays immutable. Such classes are marked
     by ZEND_ACC_IMMUTABLE flag, they are not going to be copied from opcache
     shared memory to process memory and must not be modified at all.
     Few related data structures were changed to allow addressing mutable data
     structures from immutable ones. This access is implemented through
     ZEND_MAP_PTR... abstraction macros and, basically, uses additional level of
     indirection. op_array->run_time_cache, op_array->static_variables_ptr and
     class_entry->static_members_table now have to be accessed through
     ZEND_MAP_PTR... macros.
     It's also not allowed to change op_array->reserved[] handles of immutable
     op_arrays. Instead, now you have to reserve op_array handle using
     zend_get_op_array_extension_handle() during MINIT and access its value
     using ZEND_OP_ARRAY_EXTENSION(op_array, handle).

  i. The type of the escape parameter of php_fgetcsv() and php_fputcsv() has
     been changed from char to int. This allows to pass the new constant macro
     PHP_CSV_NO_ESCAPE to this parameter, to disable PHP's proprietary escape
     mechanism.

  j. add_get_assoc_*() and add_get_index_*() are removed. Use add_assoc*(),
     add_index*() or zend_hash_*() API functions instead.

  k. Complex class declaration opcodes ZEND_ADD_INTERFACE, ZEND_ADD_TRAIT,
     ZEND_BIND_TRAITS and ZEND_VERIFY_ABSTRACT_CLASS were removed. Information
     about interfaces and traits is kept in zend_class_entry structure and
     actual linked performed by ZEND_DECLARE_...CLASS... opcode(s).
     Linked classes have ZEND_ACC_LINKED flag set.

  l. HASH_FLAG_INITIALIZED was reverted into HASH_FLAG_UNINITIALIZED. Special
     HT_IS_INITIALIZED() and HT_INVALIDATE() macro were introduced to hide
     implementation details.

  m. The write_property() object handler now returns the assigned value (after
     possible type coercions) rather than void. For extensions, it should
     usually be sufficient to return whatever was passed as the argument.

  n. Assignments to references now need to ensure that they respect property
     types that affect the reference. This means that references should no
     longer be directly assigned to, and instead a set of specialized macros of
     the form ZEND_TRY_ASSIGN* needs to be used. You can find detailed porting
     instructions as well as a compatibility shim in the wiki:
     https://wiki.php.net/rfc/typed_properties_v2#assignments_to_references

  o. ZEND_COMPILE_EXTENDED_INFO has been split into:
       ZEND_COMPILE_EXTENDED_FCALL and ZEND_COMPILE_EXTENDED_STMT
     This allows tooling to choose between profiling and debugging behaviour
     ZEND_COMPILE_EXTENDED_INFO remains and preserves behaviour.

  p. ZEND_EXT_BEGIN_FCALL is emitted after arguments are sent, this means
     that handlers may access arguments.

  q. ZEND_COMPILE_IGNORE_USER_FUNCTIONS and ZEND_COMPILE_IGNORE_INTERNAL_FUNCTIONS
     are respected by zend_get_call_op such that when set, the only possible
     call opcodes are ZEND_DO_FCALL and ZEND_DO_FCALL_BY_NAME, previously they
     were ignored by zend_get_call_op.

  r. TSRM adds tsrm_env_lock() and tsrm_env_unlock() for ZTS:
     code that may change environ and may run concurrently with user code in ZTS
     is expected to use this exclusion API to maintain as much safety as
     reasonable. This results in "thread safe" getenv/putenv in Windows and
     Unix, however functions that may read the environment without exclusion
     still exist, for example:
       - setlocale
       - mktime
       - tzset
     The above is not an exhaustive list of such functions, while getenv/putenv
     will behave as if they are safe, care should still be taken in
     multi-threaded environments.

  s. Correct support for typed properties requires the use of new macros to
     assign values to references. For more information see
     https://wiki.php.net/rfc/typed_properties_v2#impact_on_extensions.

  t. convert_to_string() and zval_get_string() are now more likely to result in
     an exception. For instructions on how to gracefully handle this see
     https://wiki.php.net/rfc/tostring_exceptions#extension_guidelines.

  u. The typedefs uint and ulong are no longer guaranteed to be available, and
     have to be replaced with standard types.

  v. Compound assignment opcodes were changed. Instead of ZEND_ASSIGN_ADD (and
     others) with 0, ZEND_ASSIGN_DIM, ZEND_ASSIGN_OBJ or
     ZEND_ASSIGN_STATIC_PROP in extended value, now we use ZEND_ASSIGN_OP,
     ZEND_ASSIGN_DIM_OP, ZEND_ASSIGN_OBJ_OP and ZEND_ASSIGN_STATIC_PROP_OP with
     ZEND_ADD (or other) in extended_value.

  w. APACHE symbol has been removed and is no longer defined.

  x. php_error_docref0() has been removed and renamed to php_error_docref().

  y. The read and write operations of php_stream_ops now return ssize_t, with
     negative values indicating an error.

========================
2. Build system changes
========================

  a. Abstract
    - The hash extension is now always available, meaning the --enable-hash
        configure argument has been removed.
    - The filter extension no longer exposes the --with-pcre-dir configure
      argument and therefore allows shared builds with ./configure for Unix
      builds.
    - Symbols HAVE_DATE, HAVE_REFLECTION, and HAVE_SPL have been removed. It
      should be considered to have these extensions always available.
    - Removed unused build time symbols: PHP_ADA_INCLUDE, PHP_ADA_LFLAGS,
      PHP_ADA_LIBS, PHP_APACHE_INCLUDE, PHP_APACHE_TARGET, PHP_FHTTPD_INCLUDE,
      PHP_FHTTPD_LIB, PHP_FHTTPD_TARGET, PHP_CFLAGS, PHP_DBASE_LIB,
      PHP_BUILD_DEBUG, PHP_GDBM_INCLUDE, PHP_IBASE_INCLUDE, PHP_IBASE_LFLAGS,
      PHP_IBASE_LIBS, PHP_IFX_INCLUDE, PHP_IFX_LFLAGS, PHP_IFX_LIBS,
      PHP_INSTALL_IT, PHP_IODBC_INCLUDE, PHP_IODBC_LFLAGS, PHP_IODBC_LIBS,
      PHP_MSQL_LFLAGS, PHP_MSQL_INCLUDE, PHP_MSQL_LFLAGS, PHP_MSQL_LIBS,
      PHP_MYSQL_INCLUDE, PHP_MYSQL_LIBS, PHP_MYSQL_TYPE, PHP_OCI8_SHARED_LIBADD,
      PHP_ORACLE_SHARED_LIBADD, PHP_ORACLE_DIR, PHP_ORACLE_VERSION,
      PHP_PGSQL_INCLUDE, PHP_PGSQL_LFLAGS, PHP_PGSQL_LIBS, PHP_SOLID_INCLUDE,
      PHP_SOLID_LIBS, PHP_EMPRESS_INCLUDE, PHP_EMPRESS_LIBS, PHP_SYBASE_INCLUDE,
      PHP_SYBASE_LFLAGS, PHP_SYBASE_LIBS, PHP_DBM_TYPE, PHP_DBM_LIB,
      PHP_LDAP_LFLAGS, PHP_LDAP_INCLUDE, PHP_LDAP_LIBS.
    - Removed unused symbols: HAVE_CURL_EASY_STRERROR, HAVE_CURL_MULTI_STRERROR,
      HAVE_MPIR, HAVE_MBSTR_CN, HAVE_MBSTR_JA, HAVE_MBSTR_KR, HAVE_MBSTR_RU,
      HAVE_MBSTR_TW.

  b. Unix build system changes
    - Added --ini-path and --ini-dir options to php-config.
    - configure --help now also outputs --program-suffix and --program-prefix
      information by using the Autoconf AC_ARG_PROGRAM macro.
    - Obsolescent macros AC_FUNC_VPRINTF and AC_FUNC_UTIME_NULL have been
      removed. Symbols HAVE_VPRINTF and HAVE_UTIME_NULL are no longer defined
      since they are not needed on the current systems.
    - Local PHP m4 unused or obsolete macros have been removed:
      PHP_TARGET_RDYNAMIC, PHP_SOLARIS_PIC_WEIRDNESS, PHP_SYS_LFS,
      PHP_AC_BROKEN_SPRINTF, PHP_EXTENSION, PHP_DECLARED_TIMEZONE,
      PHP_CHECK_TYPES, PHP_CHECK_64BIT, PHP_READDIR_R_TYPE,
      PHP_SETUP_KERBEROS.
    - Local PHP_TM_GMTOFF m4 macro replaced with Autoconf's AC_CHECK_MEMBERS.
      The HAVE_TM_GMTOFF symbol is replaced with HAVE_STRUCT_TM_TM_GMTOFF and
      HAVE_TM_ZONE symbol is replaced with HAVE_STRUCT_TM_TM_ZONE.
    - new --enable-rtld-now build option allow to switch dlopen behavior
      from RTLD_LAZY to RTLD_NOW
    - Minimum Bison version is 3.0+ for generating parser files.
    - PHP_PROG_BISON macro now takes two optional arguments - minimum required
      version and excluded versions that aren't supported.
    - PHP_PROG_RE2C is not called in the generated configure.ac for extensions
      anymore and now takes one optional argument - minimum required version.
    - with-pcre-valgrind and with-valgrind are merged, and valgrind detected by
      pkgconfig
    - Removed unused AC_PROG_CC_C_O check and the NO_MINUS_C_MINUS_O symbol.
    - Obsolescant checks for headers and functions that are part of C89 have
      been removed. The following symbols are therefore no longer defined by the
      PHP build system at the configure step and shouldn't be used anymore:
      HAVE_SETLOCALE, HAVE_LOCALECONV, HAVE_STRSTR, HAVE_STRTOL, HAVE_STRBRK,
      HAVE_PERROR, HAVE_STRFTIME, HAVE_TZNAME, HAVE_STDARG_H, HAVE_STRING_H,
      HAVE_STDLIB_H, HAVE_SYS_VARARGS_H, HAVE_ASSERT_H, HAVE_SYS_DIR_H,
      TM_IN_SYS_TIME, HAVE_STRTOD, HAVE_STRCOLL, HAVE_ERRNO_H, HAVE_MEMCPY,
      HAVE_SNPRINTF, HAVE_STDIO_H, HAVE_STRPBRK, HAVE_TIME_H, HAVE_LIMITS_H,
      HAVE_STRTOUL, HAVE_SYS_NDIR_H, HAVE_SYS_TIMES_H, PHP_HAVE_STDINT_TYPES,
      HAVE_SIGNAL_H, HAVE_STRERROR.
    - Removed unused check for dev/arandom and the HAVE_DEV_ARANDOM symbol.
    - Remove unused functions checks: HAVE_MBSINIT, HAVE_MEMPCPY, HAVE_SETPGID,
      HAVE_STRPNCPY, HAVE_STRTOULL, HAVE_VSNPRINTF, HAVE_CUSERID, HAVE_LRAND48,
      HAVE_RANDOM, HAVE_SRAND48, HAVE_SRANDOM, HAVE_STRDUP, HAVE_GCVT,
      HAVE_ISASCII, HAVE_LINK, HAVE_LOCKF, HAVE_SOCKOPT, HAVE_SETVBUF, HAVE_SIN,
      HAVE_TEMPNAM.
    - Unused check for struct cmsghdr and symbol HAVE_CMSGHDR have been removed.
    - Unused ApplicationServices/ApplicationServices.h headers check and
      HAVE_APPLICATIONSERVICES_APPLICATIONSERVICES_H symbol have been removed.
    - PHP_DEBUG_MACRO macro has been removed.
    - PHP_CHECK_CONFIGURE_OPTIONS macro has been removed. Default Autoconf's
      --enable-option-checking=fatal option can be used in the configure step
      to enable error when invalid options are used.
    - Removed unused check and symbols HAVE_SHM_MMAP_ZERO, HAVE_SHM_MMAP_FILE.
    - Removed unused check and symbol MISSING_MSGHDR_MSGFLAGS.

  c. Windows build system changes

    . Visual Studio 2019 is utilized for the Windows builds
    . Removed unused defined symbol HAVE_LIBBIND.

========================
3. Module changes
========================

  a. ext/xml
    - The public (internal) API of the ext/xml extension has been removed. All
      functions and structures are private to the extension now.

  b. ext/hash
    - The hash extension is now always available, allowing extensions to rely
      on its functionality to be available without compile time checks.