Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:85428
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain zend.com designates 209.85.160.178 as permitted sender)
MIME-Version: 1.0
Date: Tue, 24 Mar 2015 01:41:08 +0300
Message-ID: <CA+9eiLsAQPTkVuR=D0wXMFXhBNM8DQzGq0DRXzSuL+hjPsUpgA@mail.gmail.com>
To: Xinchen Hui <xinchen.h@zend.com>, Nikita Popov <nikita.ppv@gmail.com>, 
	Joe Watkins <pthreads@pthreads.org>
Cc: PHP Internals <internals@lists.php.net>
Content-Type: multipart/alternative; boundary=001a11c29cc0b2cd410511fc5cb0
Subject: Use "ropes" for string concatenation
From: dmitry@zend.com (Dmitry Stogov)

--001a11c29cc0b2cd410511fc5cb0
Content-Type: text/plain; charset=UTF-8

Hi,

Recently, Xinchen and me worked on optimization that eliminates useless
reallocations and copying during string concatenation (ZEND_ADD_STRING and
family + ZEND_CONCAT).

The idea comes from ropes, but adopted especially for our needs.
Rope is popular data structure in languages with immutable strings.
http://en.wikipedia.org/wiki/Rope_%28data_structure%29

We don't try to use ropes everywhere in the engine (at least it's too later
for 7.0), only for concatenation.

The first branch uses ropes only instead of ZEND_ADD_STRING and family.
This must be safe. The only problem is possible memory leaks on exception
(but we already have this problem anyway). The simplest way to understand
the patch - read code for new opcodes in zend_vm_def.h.

https://github.com/php/php-src/pull/1194/files

The second branch in addition uses ropes for series of ZEND_CONCAT. It
disables calls to do_operation(ZEND_CONCAT) handler of custom internal
classes.

https://github.com/php/php-src/pull/1195/files

Both make slight speed improvement (first +0.3%, second +0.6% on wordpress
home page).

We don't currently use ability to override CONCAT behavior in internal
classes, and I'm not sure if it may be useful at all. (For example Lua
doesn't provide concat meta-method). May be remove it?

Thoughts and comments are welcome.

Thanks. Dmitry.

--001a11c29cc0b2cd410511fc5cb0--