Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:73124
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain gmail.com designates 209.85.219.49 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <CAEKnhAEFCALpqA8_Gsr-TH_4GUmscrGhLRD70VcAFFV3ij68PQ@mail.gmail.com>
References: <CAEKnhAEFCALpqA8_Gsr-TH_4GUmscrGhLRD70VcAFFV3ij68PQ@mail.gmail.com>
Date: Thu, 13 Mar 2014 21:04:21 +0100
Message-ID: <CAF+90c9aPbT=CK4eCdx1FOWMHcBSr+aJmnUyJd+3Twt=At1DUQ@mail.gmail.com>
To: Jakub Zelenka <bukka@php.net>
Cc: PHP internals list <internals@lists.php.net>
Content-Type: multipart/alternative; boundary=089e015383e485f37f04f482750b
Subject: Re: [PHP-DEV] New JSON parser
From: nikita.ppv@gmail.com (Nikita Popov)

--089e015383e485f37f04f482750b
Content-Type: text/plain; charset=ISO-8859-1

On Thu, Mar 13, 2014 at 8:48 PM, Jakub Zelenka <bukka@php.net> wrote:

> Hi,
>
> I have create a new JSON parser using conditional re2c and pure pull Bison
> parser. It's a native UTF-8 parser licensed under PHP license (it can be
> used for Evil though :) ). The extension is available at
>
> https://github.com/bukka/php-jsond
>
> The encoder is taken from the current ext/json but the decoder has been
> completely rewritten.
>
> I have done some basic benchmarks (the results are on the README.md page).
> For short strings the results are almost the same (the new parser is
> usually slightly faster) but there is a big difference for longer strings.
> For example for string with 800 characters the new parser is 4 times
> faster. If the string is longer, the results are even better. It's also
> much faster for big arrays. For example decoding json_encode($_SERVER) is
> twice faster with the new parser.
>
> In addition to the speed improvements there also is a big memory saving.
> The old parser converts every serialized string to UTF-16 which requires
> additional memory. Although the memory is freed after the parsing, it can
> be problematic if the memory_limit is set. The new parser parses supplied
> string and no extra memory for conversion is allocated.
>
> I need to do more testing before creating RFC for replacing the current
> parser. There is still space for further improvements. If anyone has any
> ideas, please let me know. Or if you could test it, that would be great
> too! ;)
>
> I haven't finished the build config (it's working but it's not auto rebuild
> when you change *.y or *.re files). Need to figure out how to add
> Makefile.frag for bison and re2c . It's not working when I add
> PHP_ADD_MAKEFILE_FRAGMENT($abs_srcdir/Makefile.frag)
> $ make
> make: *** No rule to make target `jsond.lo', needed by `jsond.la'.  Stop.
> Has anyone got any idea? :)
> Also there is no build for Win atm as I don't have Windows... :( Hopefully
> someone will do it for me... :)
>

This sounds great! We solve the licensing issue *and* get a performance
benefit at the same time :)

One quick note, while looking over the code: FLOAT = INT "." UINT should be
FLOAT = INT "." DIGIT+ otherwise you'll disallow numbers with leading
zeros, e.g. 1.001. Same applies to ( INT | FLOAT ) [eE] [+-]? UINT, where
the exponent should also allow leading zeros ;)

Nikita

--089e015383e485f37f04f482750b--