Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:26299
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: error (pb1.pair.com: domain firehawksystems.com from 204.11.219.140 cause and error)
In-Reply-To: <20061103132040.GA20241@redhat.com>
References: <20061027115704.5160b509.mba2000@ioplex.com> <3F51F1DD-7B25-4302-975A-2F8E276D5EC9@firehawksystems.com> <20061103132040.GA20241@redhat.com>
Mime-Version: 1.0 (Apple Message framework v752.2)
Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed
Message-ID: <9D96A44D-3915-4526-9554-AF28A78B58F4@firehawksystems.com>
Cc: internals@lists.php.net
Content-Transfer-Encoding: 7bit
Date: Fri, 3 Nov 2006 09:04:54 -0500
To: Joe Orton <jorton@redhat.com>
Subject: Re: [PHP-DEV] What version introduced RTLD_DEEPBIND?
From: list@firehawksystems.com ("Brian J. France")

On Nov 3, 2006, at 8:20 AM, Joe Orton wrote:
> On Fri, Oct 27, 2006 at 12:12:46PM -0400, Brian J. France wrote:
>> and I plan on removing it from our internal builds as it causes more
>> problems than I think it fixes (on RHEL only).
>
> Hi, what problems do you see from the use of RTLD_DEEPBIND?

Where to start? :)

First issue we have is we build our extension linked with a static  
library called apstubs.  apstubs is a hack where all exported apache  
functions, structs, etc are defined weak (function print a error and  
abort).  We do this so PHP extensions that are mainly for apache  
only, can still load on the command line without issue (assuming weak  
code path is not executed).  When things are run in apache the strong  
symbols from apache override the weak ones in the library.

With RTLD_DEEPBIND we don't get this behavior because that flag and  
force it to use the weak symbols even when the strong symbols are  
present.  I understand the reasoning for this behavior (not being  
able to namesapce and having duplicate symbols), but we rely on weak/ 
strong symbols in a lot of our code and this just causes more problems.

A second issue we have is when RTLD_DEEPBIND is defined the rtld  
can't find symbols in a already loaded lib.  This one is strange and  
I can't figure it out, but if I had to guess I would be it is a  
problem with libraries not getting unloaded between the first and  
second pass of apache startup (which we are trying to track down now,  
expect a new or updated RH bug soon if we find something).  But for  
some reason turning off RTLD_DEEPBIND fixes the core dump (not the  
unloading issue, our FreeBSD kernel team is looking into it to see if  
they can help).

You can see this issue with RH bug 84372.  Look at the comment on  
"08-30-2006 01:40pm" on how to build open source software and make it  
happen.  I am going to see if removing RTLD_DEEPBIND fixes this issue  
as well.

Brian

  - bfrance posting as mason on the RH issue tracker