Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:90995
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: error (pb1.pair.com: domain belski.net from 85.214.73.107 cause and error)
To: <internals@lists.php.net>
Cc: "'Pierre Joye'" <pierre.php@gmail.com>,
	"'Dmitry Stogov'" <dmitry@zend.com>,
	"'Xinchen Hui'" <laruence@gmail.com>,
	"'Nikita Popov'" <nikita.ppv@gmail.com>
Date: Thu, 28 Jan 2016 15:48:42 +0100
Message-ID: <03ca01d159da$fe2d76b0$fa886410$@belski.net>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: 7bit
Thread-Index: AdFZ12y7QFDTZgpOSVa5W4LeeaSTKQ==
Content-Language: en-us
Subject: Streams and I/O refactoring approach
From: anatol.php@belski.net ("Anatol Belski")

Hi,

I'm writing to ask for a review on the following:

https://gist.github.com/weltling/29779b61db26c62b5ab0

It is quite far from being even near to be ready. It is an approach for the
streams refactoring. After some time playing around, I think that the ideas
I was initially approaching are to the big part not applicable to our code
base in its current state. For one, "isolate the actual concrete low level
implementations by platforms" - it's probably the goal 1. that has to be
done before anything else. However creating an isolated layer of structs for
every platform (libuv alike) most likely is not doable at the current stage,
if at all (and whether makes sense).

So in this patch, I've started to create an I/O layer in win32, with the
goal to create a framework global to the whole PHP. This serves as the first
unavoidable step for the separation by platform. Ideally, we should have all
the abstracted low level I/O code for platforms concentrated at one place,
while currently all possible code parts of win32 I/O are spread over all the
code base - Zend, TSRM, ext/standard, win32.

In the current state it is hard to both maintain and to develop these pieces
of our codebase. So sounds like a gross refactoring which might last for
some longer time. For the Windows part it'll bring a huge functional
advantage of using the native win32 APIs which means support for multibyte
and long filepath and improved security, later probably overlapped I/O. For
the overall codebase it'll mean simpler and better readable code, so more
stability and possibility to extend with new modern features more easily.

Practically in this patch, with the approach I've described, I've
implemented several helpers and routines replacements for Windows that use
the native APIs. Particularly PHP's fopen and mkdir there support multibyte
filepaths and implementing the long path support is then quite easy. For
that to work cross platform, I define macros like php_ioutil_open,
php_ioutil_mkdir, etc. for the low level POSIX surrogates. There, I used
several parts from libuv to not to do everything from scratch. Also I've
started to pull the win32 related code out from zend_virtual_cwd.c, but
that's just a start. The plus of such approach is also that the work can be
done incrementally, moving the functions one by one, as the unported code
will stay compatible with function signatures and functionality.

Please let me know what you think about this approach as it's a very early
stage which would be good to take the corrections.

Thanks

Anatol