rfc:new-output-api

Request for Comments: New Output API

Introduction

Quoting an old rant:

See http://marc.info/?l=php-internals&m=114104110826804&w=2

The current implementation uses way too less commands to do too many different operations, which almost always causes problems with output handlers that need to maintain a state or context.

Short example:

ob_clean() causes havoc with ext/zlib ob handlers and any handlers found in pecl/http. ob_clean() passes currently buffered input through the handler and discards it, which means in the case of the gz handler that the gzip/zlib header bytes (and more) are missing from the sent data.

Patches

Original patches may exist at http://dev.iworks.at/PATCHES. HEADs implementation probably still exists in branches/FIRST_UNICODE_IMPLEMENTATION. Jani ported it to PHP_5_3, which probably still exists in branches/PHP_5_4.

README

Quoting the original README:

$Id: README.NEW-OUTPUT-API 219039 2006-08-30 07:39:09Z mike $


API adjustment to the old output control code:

	Everything now resides beneath the php_output namespace,
	and there's an API call for every output handler op.

	Checking output control layers status:
		// Using OG()
		php_output_get_status(TSRMLS_C);

	Starting the default output handler:
		// php_start_ob_buffer(NULL, 0, 1 TSRMLS_CC);
		php_output_start_default(TSRMLS_C);

	Starting an user handler by zval:
		// php_start_ob_buffer(zhandler, chunk_size, erase TSRMLS_CC);
		php_output_start_user(zhandler, chunk_size, flags TSRMLS_CC);

	Starting an internal handler whithout context:
		// php_ob_set_internal_handler(my_php_output_handler_func_t, buffer_size, "output handler name", erase TSRMLS_CC);
		php_output_start_internal(handler_name_zval, my_php_output_handler_func_t, chunk_size, flags TSRMLS_CC);

	Starting an internal handler with context:
		// not possible with old API
		php_output_handler *h;
		h = php_output_handler_create_internal(handler_name_zval, my_php_output_handler_context_func_t, chunk_size, flags TSRMLS_CC);
		php_output_handler_set_context(h, my_context, my_context_dtor);
		php_output_handler_start(h TSRMLS_CC);

	Testing whether a certain output handler has already been started:
		// php_ob_handler_used("output handler name" TSRMLS_CC);
		php_output_handler_started(handler_name_zval TSRMLS_CC);

	Flushing one output buffer:
		// php_ob_end_buffer(1, 1 TSRMLS_CC);
		php_output_flush(TSRMLS_C);

	Flushing all output buffers:
		// not possible with old API
		php_output_flush_all(TSRMLS_C);

	Cleaning one output buffer:
		// php_ob_end_buffer(0, 1 TSRMLS_CC);
		php_output_clean(TSRMLS_C);

	Cleaning all output buffers:
		// not possible with old API
		php_output_clean_all(TSRMLS_C);

	Discarding one output buffer:
		// php_ob_end_buffer(0, 0 TSRMLS_CC);
		php_output_discard(TSRMLS_C);

	Discarding all output buffers:
		// php_ob_end_buffers(0 TSRMLS_CC);
		php_output_discard_all(TSRMLS_C);

	Stopping (and dropping) one output buffer:
		// php_ob_end_buffer(1, 0 TSRMLS_CC)
		php_output_end(TSRMLS_C);

	Stopping (and dropping) all output buffers:
		// php_ob_end_buffers(1, 0 TSRMLS_CC);
		php_output_end_all(TSRMLS_C);

	Retrieving output buffers contents:
		// php_ob_get_buffer(zstring TSRMLS_CC);
		php_output_get_contents(zstring TSRMLS_CC);

	Retrieving output buffers length:
		// php_ob_get_length(zlength TSRMLS_CC);
		php_output_get_length(zlength TSRMLS_CC);

	Retrieving output buffering level:
		// OG(nesting_level);
		php_output_get_level(TSRMLS_C);

	Issue a warning because of an output handler conflict:
		// php_ob_init_conflict("to be started handler name", "to be tested if already started handler name" TSRMLS_CC);
		php_output_handler_conflict(new_handler_name_zval, set_handler_name_zval TSRMLS_CC);

	Registering a conflict checking function, which will be checked prior starting the handler:
		// not possible with old API, unless hardcoding into output.c
		php_output_handler_conflict_register(handler_name_zval, my_php_output_handler_conflict_check_t TSRMLS_CC);

	Registering a reverse conflict checking function, which will be checked prior starting the specified foreign handler:
		// not possible with old API
		php_output_handler_reverse_conflict_register(foreign_handler_name_zval, my_php_output_handler_conflict_check_t TSRMLS_CC);

	Facilitating a context from within an output handler callable with ob_start():
		// not possible with old API
		php_output_handler_hook(PHP_OUTPUT_HANDLER_HOOK_GET_OPAQ, (void *) &custom_ctx_ptr_ptr TSRMLS_CC);

	Disabling of the output handler by itself:
		//not possible with old API
		php_output_handler_hook(PHP_OUTPUT_HANDLER_HOOK_DISABLE, NULL TSRMLS_CC);

	Marking an output handler immutable by itself because of irreversibility of its operation:
		// not possible with old API
		php_output_handler_hook(PHP_OUTPUT_HANDLER_HOOK_IMMUTABLE, NULL TSRMLS_CC);

	Restarting the output handler because of a CLEAN operation:
		// not possible with old API
		if (flags & PHP_OUTPUT_HANDLER_CLEAN) { ... }

	Recognizing by the output handler itself if it gets discarded:
		// not possible with old API
		if ((flags & PHP_OUTPUT_HANDLER_CLEAN) && (flags & PHP_OUTPUT_HANDLER_FINAL)) { ... }


Output handler hooks

	The output handler can change its abilities at runtime. Eg. the gz handler can
	remove the CLEANABLE and REMOVABLE bits when the first output has passed through it;
	or handlers implemented in C to be used with ob_start() can contain a non-global
	context:
		PHP_OUTPUT_HANDLER_HOOK_GET_OPAQ
			pass a void*** pointer as second arg to receive the address of a pointer
			pointer to the opaque field of the output handler context
		PHP_OUTPUT_HANDLER_HOOK_GET_FLAGS
			pass a int* pointer as second arg to receive the flags set for the output handler
		PHP_OUTPUT_HANDLER_HOOK_GET_LEVEL
			pass a int* pointer as second arg to receive the level of this output handler
			(starts with 0)
		PHP_OUTPUT_HANDLER_HOOK_IMMUTABLE
			the second arg is ignored; marks the output handler to be neither cleanable
			nor removable
		PHP_OUTPUT_HANDLER_HOOK_DISABLE
			the second arg is ignored; marks the output handler as disabled


Open questions

	Should the userland API be adjusted and unified?

	Many bits of the manual (and very first implementation) do not comply
	with the behaviour of the current (to be obsoleted) code, thus should
	the manual or the behaviour be adjusted?

Comments

  • BC note: This [intentionally] changes ob_start() behavior, where a chunk_size parameter value of 1 no longer sets chunk_size to 4096. Ref: Bug #46903

Changelog

  • 2010-04-07: created
  • 2010-04-07: added comments section, and ob_start() comment
rfc/new-output-api.txt · Last modified: 2013/10/29 18:29 by nikic