My original post didn't make it to the openssl-users@openssl.org users
list, but I think I've found the problem now: In OpenSSL 0.9.8a,
_FILE_OFFSET_BITS is defined in bss_file.c. This is a very strange place:
It means that all other units will not see this definition. If I configure
with:
./Configure -D_FILE_OFFSET_BITS=64 linux-elf
...the problem goes away: Both the openssl tool and rdesktop now correctly
passes the off_t_problem.py test.
So, I think _FILE_OFFSET_BITS should not be defined in bss_file.c but
rather somewhere else.
Regards,
Peter Åstrand
On Fri, 24 Mar 2006, Peter Åstrand wrote:
I've been trying to build a version of rdesktop (www.rdesktop.org) with
largefile support and verify it with off_t_problem.py
(http://ac-archive.sourceforge.net/largefile/pythonscript.html). Despite my
efforts, I always gets:
weirdos: /home/peter/thin/rdesktop/rdesktop/rdesktop IMPORTS open64/open
fopen64/fopen
WARNING: importing both 32bit and 64bit off_t symbols is very very dangerous!
After some more investigations, I've found out that this is true for the
"openssl" binary as well!:
imports: /home/peter/openssl-maggie/bin/openssl 32++ open.. fopen.. lseek..
readdir..
imports: /home/peter/openssl-maggie/bin/openssl ++64 fopen64
weirdos: /home/peter/openssl-maggie/bin/openssl IMPORTS fopen64/fopen
WARNING: importing both 32bit and 64bit off_t symbols is very very dangerous!
summary: found 1 weirdos - too dangerous to use them: (file bug report!)
summary: found 0 badlinks to be checked closer ( 0 are clean)
This happens both on RedHat 7.3 and Fedora Core 4. I'm using OpenSSL 0.9.8a.
So, what's broken:
1) off_t_problem.py
or
2) OpenSSL
or
3) The operating system
?
Regards,
--
Peter Åstrand ThinLinc Chief Developer
Cendio http://www.cendio.se
Teknikringen 3
583 30 Linköping Phone: +46-13-21 46 00