Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:37490
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: error (pb1.pair.com: domain lerdorf.com from 204.11.219.139 cause and error)
Message-ID: <481F8BCD.30207@lerdorf.com>
Date: Mon, 05 May 2008 15:35:57 -0700
User-Agent: Thunderbird 2.0.0.4 (Macintosh/20070604)
MIME-Version: 1.0
To: Matt Wilmas <php_lists@realplain.com>
CC:  internals@lists.php.net
References: <013a01c8aec5$b189d650$0201a8c0@pc1>
In-Reply-To: <013a01c8aec5$b189d650$0201a8c0@pc1>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Subject: Re: [PHP-DEV] [PATCH] Some string changes/optimizations
From: rasmus@lerdorf.com (Rasmus Lerdorf)

A bit of a messy patch in that it doesn't have a single top-level 
directory, but after hacking it, it applied.  For others testing this, 
make sure you run ext/tokenizer/tokenizer_data_gen.sh and of course 
genfiles with the latest re2c.

I'm still not getting something that works though.  Everything compiles 
fine, but make test is throwing:

Fatal error: Invalid opcode 56/16/8. in 
/Users/rasmus/php53/run-tests.php on line 547

Could we get a cleaner patch to have a look at?  One that doesn't hurt 
my brain quite as much to get working?

-Rasmus



Matt Wilmas wrote:
> Hi all,
> 
> This patch...  Saves one opcode per interpolated string by moving what was
> the first ADD_STRING/ADD_VAR op into INIT_STRING; changes the
> ADD_[CHAR|STRING|VAR] op types from (TMP, <whatever>) to (<whatever>,
> UNUSED), since that seems to make more sense (see ADD_ARRAY_ELEMENT) and
> simplifies things a bit, and saves some znode copying while compiling.
> 
> Next, after nowdoc support was added, I noticed it was mostly duplicate
> heredoc code in the scanner, so I combined them for the most part, removing
> the NOWDOC tokens, etc.  Is that OK?  It seems like they may have just been
> there for the parser... but I updated the parser so that static heredocs
> also work like nowdocs.  Also removed the ST_START_[HEREDOC|NOWDOC] states
> in the scanner, by doing their work (to catch immediate ending label) in the
> initial heredoc rule.
> 
> While removing the NOWDOC references from zend_highlight.c, I made a little
> change that I think improves (err... adds) highlighting of variables in
> double-quoted strings, and makes literal text in heredocs/backticks the
> correct "highlight_string" color.  This makes coloring consistent across all
> strings. :-)  Before/after example: http://realplain.com/php/highlight.html
> 
> Speaking of backticks, while updating their parser grammar, they now behave
> more precisely like shell_exec(), in that `cmd` (constant string) or `$cmd`
> (one variable) won't use INIT_STRING and create a temporary variable...
> 
> Oh, almost forgot a couple things -- for HEAD: updated the yyless()
> definition to match 5_3 (Nuno didn't get to it yet...), and moved
> HANDLE_NEWLINES() so nowdoc text is copied first, otherwise the line number
> would be off if there's an error in zend_copy_scanner_string.  For 5_3:
> there are NOWDOC references in tokenizer_data.c that the patch doesn't
> remove, as I think it's supposed to be automated, though they're missing in
> HEAD.
> 
> http://realplain.com/php/string_optimizations.diff
> http://realplain.com/php/string_optimizations_5_3.diff
> 
> 
> Thanks,
> Matt
> 
> P.S. (for Marcus?), the ({LABEL}|([']{LABEL}['])|(["]{LABEL}["])) part of
> the heredoc rule crashed re2c (Windows) until I added the extra ( ), which
> shouldn't be needed AFAIK and was fine with just 2 alternate matches...
> 
>