Whoops - didn't mean to submit that yet.



I can't reproduce with my test install on RHEL 4, although there are 
subtle differences there (beyond newer PHP and Apache version), like 
the PHP and Apache configs, different version of PEAR modules, etc.



But I will chalk it up to a broken PHP 4.3.2.  Need to upgrade to RHEL 
4 anyway, so this is just more motivation to do it.

Yes, I am well aware that .  Its a bug in Apache or PHP, but I don't 
know enough about why the routines in Horde trigger it, to file an 
intelligent bug report with either of those entities.



FWIW, I threw together another test site on RHEL 4, which uses Apache 
2.0.52 and PHP 4.3.9.

There is really no way this can be a bug in Horde (where the "error" 
is, btw - not IMP) or IMP.

I agree it doesn't make any sense but I'm not currently running a 
cache.  My Apache periodically segfaults with eaccelerator but thats 
another issue to track down another day :)

The error message makes no sense and is rather a sign of a badly 
broken PHP or bytecode cache.

I funnel all my log summaries to a folder.  The one daily summary from 
a Samba server generally contains a few lines of garbage, from the 
Samba daemon.  Somedays IMP displays the message, other days, it won't 
display it.  Today I set out to track down why.



Apache is 2.0.46 and PHP 4.3.2 (Red Hat Enterprise Linux 3, Update 4)



The logs contain the following (the Invalid key 0 error is from Samba):



[29-Apr-2005 20:16:40] PHP Notice:  Undefined variable:  )  Invalid 
key 0 given to dptr_close^_<8b>^H^H<84>RkB^M

......contents of most of the rest of the mail message

in /horde/lib/Horde/MIME/Viewer.php on line 419



Just preceeding the line in the message that is logged is the garbage 
line that triggers this:



½¶«Y^OJÂÎ~ge %OÉ*þÇ~[7¤ùÆ^A&^A

s^As]é½°

^O¸¸¸¸¸¸?tKQ½~C+v%û©¸¸¸_u%{~Y:×~K\Í?~I^L]É^ñ¿»O`Ñw      h9

&~\R~NÑ^AÐÆ¡U~E2!~W\=bh¤õ£ÎJÆö;~@«~[~E5zý^D^Dw3k^GÑÃö¸«¿»^U¯^X'ï®5m1Ec~[ÑË>Å~C^VªQ"~J~L~RãÒþ^\ø:¡^SÆNACç^CÐ^\.!¤q~B³wNËh|~RÍ| ^Uø)n*F~KJ^SLþ©âU%^\ãß«/E 
ç~Xº^A



So what appears to be happening is during the parsing of the message, 
IMP or a Horde class hits the garbage line.  And everything after that 
gets passed to a function - _getIcon??? and it chokes on that.



Uses up all available memory and Apache segfaults.  But why one line 
in a message is triggering this in _getIcon is beyond me.  How do I 
isolate what exactly is causing this?



Thanks,

Kevin