Hi Luk,
Can you take a look at src/config.h after the failed build to figure out
which HAVE_XXX_OS is defined? Also, the definition of ioctl() in the
header files?
The discussion and details are below.
Thanks,
-- John
--- Begin Message ---
A NOTE has been added to this issue.
======================================================================
http://bugs.bacula.org/view.php?id=1359
======================================================================
Reported By: jgoerzen
Assigned To:
======================================================================
Project: bacula
Issue ID: 1359
Category: Storage Daemon
Reproducibility: always
Severity: major
Priority: normal
Status: feedback
======================================================================
Date Submitted: 2009-08-28 22:52 BST
Last Modified: 2009-08-29 15:35 BST
======================================================================
Summary: Doesn't compile on Alpha
Description:
During the attempt to autobuild bacula 3 on Debian, this error was
reported:
make[1]: Entering directory
`/build/buildd-bacula_3.0.2-3-alpha-1PG1hy/bacula-3.0.2/debian/tmp-build-sqlite3/src/stored'
Compiling dev.c
dev.c: In member function 'void DEVICE::init_backend()':
dev.c:307: error: invalid conversion from 'int (*)(int, long unsigned int,
...)throw ()' to 'int (*)(int, int, ...)'
make[1]: *** [dev.o] Error 1
make[1]: Leaving directory
`/build/buildd-bacula_3.0.2-3-alpha-1PG1hy/bacula-3.0.2/debian/tmp-build-sqlite3/src/stored'
make: *** [build-stamp-sqlite3] Error 2
The full error log is available at
https://buildd.debian.org/fetch.cgi?pkg=bacula;ver=3.0.2-3;arch=alpha;stamp=1251420917
The line of source in question is
d_ioctl = ::ioctl;
Version 2.4.4 successfully built on this platform.
======================================================================
----------------------------------------------------------------------
(0004556) ebollengier (administrator) - 2009-08-29 07:47
http://bugs.bacula.org/view.php?id=1359#c4556
----------------------------------------------------------------------
We have a specific define for Alpha that maps ioctl argument to int. Does
it mean that it should be mapped to long unsigned int ? Or it's a different
architecture ?
#ifdef __alpha__
#define OSF 1
#undef ioctl_req_t
#define ioctl_req_t int
#endif
----------------------------------------------------------------------
(0004557) kern (administrator) - 2009-08-29 09:17
http://bugs.bacula.org/view.php?id=1359#c4557
----------------------------------------------------------------------
I don't understand why vendors don't simply follow the POSIX standard.
Eric, for Alpha (and a few other that don't respect standards), the second
argument of an ioctl call is "unsigned int" rather than "int".
The definition you show above should be:
#define ioctl_req_t unsigned int
However, who knows if it is going to break something somewhere else.
----------------------------------------------------------------------
(0004562) jgoerzen (reporter) - 2009-08-29 14:59
http://bugs.bacula.org/view.php?id=1359#c4562
----------------------------------------------------------------------
It appears that your #ifdef __alpha__ is for HP Tru64 UNIX (previously
known as OSF/1) running on that platform. Linux also runs on that
platform, and may have different semantics.
Kern, my reading of the error suggests that it should be
#define ioctl_req_t long unsigned int
but I may be missing something.
----------------------------------------------------------------------
(0004563) jgoerzen (reporter) - 2009-08-29 15:02
http://bugs.bacula.org/view.php?id=1359#c4563
----------------------------------------------------------------------
I should clarify. It is invalid to infer what OS is being used from what
architecture is in use. You can't assume that an Alpha is running OSF/1,
as Debian, NetBSD, etc. all run on it as well.
----------------------------------------------------------------------
(0004565) kern (administrator) - 2009-08-29 15:28
http://bugs.bacula.org/view.php?id=1359#c4565
----------------------------------------------------------------------
Well, when we are compiling Bacula, we *must* know what OS is running.
Perhaps it was a mistake for the author to use __alpha__ rather than the
OS. This code was submitted by a user, so I don't exactly know what
OS/platform it was for.
Perhaps if you have the system in front of you, you can figure out what
the correct #ifdefing should be. In general, it should be something like:
#ifdef HAVE_xxx_OS
...
...
#endif
I suspect that we should leave the current __alpha__ #ifdefing alone and
just add a new one for your case after it.
What is actually used for ioctl_req_t varies a bit, but yes, you are right
in the case of the error you were seeing (and for most of the silly
non-POSIX compatible OSes) it should be:
#define ioctl_req_t unsigned long
For the system in question, I suggest you look in the system header files
for what is really defined rather than relying on the output from g++.
----------------------------------------------------------------------
(0004568) kern (administrator) - 2009-08-29 15:35
http://bugs.bacula.org/view.php?id=1359#c4568
----------------------------------------------------------------------
I forgot to mention that you can look in <bacula>/src/config.h after
running ./configure to know which HAVE_XXX_OS is defined.
Issue History
Date Modified Username Field Change
======================================================================
2009-08-28 22:52 jgoerzen New Issue
2009-08-29 07:47 ebollengier Note Added: 0004556
2009-08-29 07:47 ebollengier Status new => feedback
2009-08-29 09:17 kern Note Added: 0004557
2009-08-29 14:59 jgoerzen Note Added: 0004562
2009-08-29 15:02 jgoerzen Note Added: 0004563
2009-08-29 15:28 kern Note Added: 0004565
2009-08-29 15:35 kern Note Added: 0004568
======================================================================
--- End Message ---
