Log excerpt from discussion with upstream about this issue

php-irc-2008-08-07-excerpt.log (text/plain), 6.88 KB, created by Christian Hoffmann (RETIRED) on 2009-04-24 09:11:39 UTC

**** BEGIN LOGGING AT Thu Aug  7 01:27:10 2008

14:58:41 <hoffie> bleh.. are there any tips on tracking down heisenbugs? :p
14:58:44 <hoffie> Zend/tests/bug31177.php crashes here
14:59:04 <@tony2001> hoffie: which GCC ?
14:59:09 <hoffie> as soon as disabling gcc optimizer (-O0) or adding debug flags (-ggdb3 or --enable-debug) it goes away
14:59:14 <hoffie> gcc version 4.3.1 (Gentoo 4.3.1-r1 p1.1) 
14:59:20 <@tony2001> yeah, I've seen that either
14:59:20 <hoffie> and it seems to be 4.3-specific
14:59:38 <hoffie> 5.2.6 btw
14:59:40 <@tony2001> only with GCC 4.3.x and only with prerelease/rc versios
15:00:07 <@tony2001> upgrading/downgrading GCC fixes it
15:00:08 <hoffie> well, i'm using the final version, not a pre-release as far as i can see
15:00:25 <hoffie> hm i've got 4.2.4 as well, will try that
15:01:15 <hoffie> what is this anyway? is it more likely a bug in the gcc optimizer or a bug in php which only becomes apparent in this environment?
15:01:58 <@tony2001> I believe it's an optimizer bug, since it only shows with certain GCC version and certain optimization level
15:03:00 <hoffie> yeah, 4.2.4 works without problems...
15:03:30 <hoffie> sucks.. we'll see how many real world applications are affected by this
15:05:06 <hoffie> there was a similar discussion on -internals some days/weeks ago, wasnt there? regarding a gcc pre-release on fedora core i think.. and there was a patch?
15:05:32 <@tony2001> hmm.. if there was, I didn't see it
15:06:04 <@ScottMac> was a bug with gcc 4.2 pre-release
15:06:46 <@ScottMac> http://bugs.php.net/bug.php?id=43487
15:07:35 <@ScottMac> http://cvs.php.net/viewvc.cgi/ZendEngine2/zend_strtod.c?r1=1.37&r2=1.38
15:08:19 <hoffie> hm, i think i've already seen this in the past, but i don't really understand the "fix" (or rather workaround?)
15:08:32 --> meus_!~[e-mail]
15:08:33 <hoffie> to me it looks like arbitrarily adding "volatile" will fix this?
15:10:11 <@ScottMac> hoffie, volatile tells the compiler something funky might happen to this var
15:11:31 <@ScottMac> without it the compiler might just re-use the value rather than load it from the registers again
15:11:35 <hoffie> well, in my case i don't even know which variable it is..
15:11:49 <@ScottMac> but it makes it less optimal
15:11:53 <@ScottMac> and hence slower
15:12:01 <@ScottMac> so if its a compiler bug its best to make that compiler fail
15:12:02 <hoffie> well, i prefer slower over crashing
15:12:26 <antab> volatile tells the compiler to fetch the data every time it has to use it. not load it to a register and keep it there.
15:13:02 <@Pierre> antab: depends
15:13:09 <@Pierre> but it is everything you want but portable
15:13:19 <@ScottMac> mainly for pointers that references a hardware register
15:13:42 <antab> dosen't impackt portablity, only compiler optimizisions
15:13:58 <@Pierre> antab: it does impact portability.
15:13:59 <antab> unless the compiler dosen't support volatile:P
15:14:08 <@Pierre> and it may be not TS 
15:14:29 <@Pierre> (if use with static)
15:14:55 --> zoe_!~[e-mail]
15:15:15 <@Pierre> there is even a cert doc describing how to do not use it =)
15:22:25 <antab> is there any buildin or extension to allow me to see all memory allocated in PHP and from there in my script it was allocated?
15:22:47 -!- Kalle MODE #php.pecl +oo felipe_ Rasmus
15:22:52 <@tony2001> I don't think it's possible even theoretically
15:24:05 <lsmith_> tony2001: thx!
15:24:22 <lsmith_> Pierre: is the htscanner stuff on your radar .. or should i poke Jani about it?
15:24:23 <antab> Its not much of a problem to implement, would slow the execution down alot but i dont care about the speed when debugging.
15:24:26 <@tony2001> you're welcome
15:25:24 <@Pierre> lsmith_: it is
15:25:53 <@Pierre> lsmith_: but the question remains 
15:27:28 <lsmith_> which one? if todo this in a generic fashion or not?
15:28:04 -!- NICK lsmith_ -> lsmith
15:28:11 --> auroraeos!~[e-mail]
15:32:39 <antab> hmm. xdebug should show up in phpinfo() right?
15:33:05 <@Derick> yes, twice even
15:34:02 <antab> funky, dosent show up and no errors in the log about unable to open it:/ but i do get a error if i change the zend_extension to extension (as i assumed)
15:35:25 <@Rasmus> display_startup_errors=On
15:35:36 <@Rasmus> probably an API version mismatch
15:37:07 <antab> already enabled. self compiled 5.2.3 with xdebug 2.0.3
15:38:31 <antab> i guess there isin't alot of error checks around zend_extension=. added a typo to the name and still no errors:/
15:40:37 <@kore_> zend_extension(_debug)(_ts)
15:41:22 <@tony2001> yeah, the directive is different for debug and ts builds
15:41:36 * lsmith pokes rza
15:41:48 <-- bfrance!~[e-mail] quit (Quit: bfrance)
15:42:06 --> bfrance!~[e-mail]
15:45:51 <@bjori> http://trac.symfony-project.org/browser/trunk/lib/cache/sfAPCCache.class.php?rev=6365
15:45:59 <@bjori> am I missing the joke somewhere?
15:47:29 <lsmith> bjori: ?
15:47:53 <@tony2001> ?
15:48:07 <@bjori> class wrapping all the apc functions.. with oneliners
15:48:10 <@Derick> we should get rid of that no warnings on zend-ext crap - and the 4 different settings
15:48:13 <@Derick> it's quite annoying
15:48:25 <lsmith> bjori: the idea is a unified API
15:48:28 <lsmith> for caching layers
15:48:29 <@tony2001> bjori: that's the way most of the frameworks work
15:48:32 <@Derick> luckily, xdebug won't be a true "zend ext" with 2.1
15:49:00 <@bjori> frameworks suck
15:49:03 <@bjori> enterprises suck
15:49:05 <@bjori> wikis suck
15:49:12 <@ScottMac> helgi sucks
15:49:14 <@tony2001> life?
15:49:27 <@Derick> antab: getting anywhere?
15:49:47 <@bjori> tony2001: ..without frameworks, enterprises and wikis? perfect! :D
15:49:57 <@tony2001> hah
15:50:03 <@bjori> Derick: using that "stealth hack"?
15:50:25 <lsmith> gna .. now i have to try and talk to sql server from linux
15:51:15 <@Derick> bjori: yes
15:51:19 -!- felipe_ MODE #php.pecl +oooo auroraeos helly25 lsmith zoe_
15:53:07 --> rrichards!~rrichards@67.158.171.203
15:53:34 -!- felipe_ MODE #php.pecl +o rrichards
15:56:04 --> Frol!~[e-mail]
15:57:11 <rza> Derick: still no response?
15:57:44 <rza> lsmith: ?
15:58:09 <@Derick> no
15:59:04 --> preinheim!~[e-mail]
16:01:44 <@bjori> Derick: did you ever get reply from wez regarding doc-all@?
16:01:59 -!- NICK _philip__ -> philip__
16:02:14 <@Derick> bjori: negative
16:02:20 <@bjori> :(
16:02:37 <@tony2001> what is doc-all@ ?
16:02:57 <@bjori> hopefully an alias to all the doc lists
16:03:15 <@bjori> not my favorite task to copy&paste 30lists when breaking stuff :)
16:03:40 <@tony2001> ah.. well, add an entry to your address book then =)
16:04:20 <@bjori> I'm not that smart :P
16:05:09 <@tony2001> lazy bstard =)
16:08:51 <-- lsmith!~lsmith@195.226.16.50 quit (Quit: lsmith)
16:09:29 <antab> Derick, will debug it more tonight
16:09:41 <@Derick> k