Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:13372
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Message-ID: <5.1.0.14.2.20041018104130.06e940b0@localhost>
Date: Mon, 18 Oct 2004 11:14:20 +0200
To: "Kamesh Jayachandran" <kameshj@fastmail.fm>
Cc: "Kamesh Jayachandran" <kameshj@fastmail.fm>,internals@lists.php.net,
 zeev@php.net,helly@php.net,andi@php.net
In-Reply-To: <1098082359.21967.206675358@webmail.messagingengine.com>
References: <1097215824.21523.206031834@webmail.messagingengine.com>
 <1097215824.21523.206031834@webmail.messagingengine.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Subject: Re: [PHP-DEV] why to call php_output_activate(TSRMLS_C) from
  php_module_startup(reposting first time)(bug #30397)
From: zeev@zend.com (Zeev Suraski)

At 08:52 18/10/2004, Kamesh Jayachandran wrote:
>Can someone respoond to this.
>
>With regards
>Kamesh Jayachandran
>On Thu, 07 Oct 2004 23:10:24 -0700, "Kamesh Jayachandran"
><kameshj@fastmail.fm> said:
> > Hi All,
> > I could see php_output_activate(TSRMLS_C) getting called from
> > php_module_startup immedeately after php_output_startup.
> >
> > Why is this needed? As each SAPI modules call
> > php_output_activate(TSRMLS_C) explicitly as a part of request startup.
> >
> > Why am I concerned about this?
> >
> > Ans: php_output_activate(TSRMLS_C) sets
> > OG(php_body_write) = php_ub_body_write;
> > which causes calls to php_printf to use php_ub_body_write which depends
> > on SG(request_info) which is null while php_printf is invoked as part of
> > apache startup error(Like Invalid extension/ Non existent extension)
> > logging.
> >
> > This causes segmentation fault.
> >
> > This change seems to have been introduced in version 1.371 of
> > main/main.c by Zeev(3 years and 3 months ago).
> > The comment says,
> > Fix a major thread safety bug in the output mechanism
> > @- Fixed a major memory corruption bug in the thread safe version (Zeev)
> > Can I remove a call to php_output_activate(TSRMLS_C) from
> > php_module_startup?
> >

Actually this existed beforehand, the change just shifted code around (the 
point of this commit was actually NOT to call php_output_startup() on every 
request startup, and it introduced php_output_activate() which is the one 
being called on every request startup instead).
 From looking at the source code it appears that the call to 
php_output_activate() during php_module_startup() was added to keep 
'compatibility' with the code prior to this commit.  It looks that this 
commit fixed the main problem but kept a minor one.  I *think* it can be 
safely removed.

> > PS:What is the use of display_startup_errors? What does display mean
> > with respect to startup as it might not have any associated request
> > context?

It has to do with errors that happen during request startup (such as file 
upload related errors) which cannot be trapped in userland.  The check 
currently present php_error_cb() appears to be messy, though - it should 
probably be fixed to be just:
if (PG(display_errors)
         && module_initialized
         && (!PG(during_request_startup) || PG(display_startup_errors)) {

This messiness may be related to the other issue you mentioned.

Zeev