The ongoing saga of lsof-4.71.1

[Michael Edenfield]

I am one of the people having problems building lsof's latest release on
-CURRENT and have spent a few hours trying to figure out why.  I managed
to get the program to build but I can't figure out what went wrong in
the first place, or more specifically, why *everyone* isn't having this
problem.

My ports tree was cvsup's immediately before I began.  I also deleted
/usr/includes and did a `make buildincludes;  make installincludes` in
/usr/src before beginning.

The error, as usual, was in machine.h in the lsof work directory.  This
is a symlink to dialiects/freebsd/machine.h and contains the following
code:

#include <sys/types.h>

#if     FREEBSDV>=520
/*
 * In FreeBSD >= 5.2 the cpumask_t typedef is only made in <sys/types.h> if
 * _KERNEL is predefined.  However, predefining _KERNEL before #include'ing
 * <sys/types.h> causes redefinition errors for boolean_t and vm_page_t when
 * <vm/vm.h> is #include'd with _KERNEL predefined.  Since lsof must have
 * _KERNEL predefined when <vm/vm.h> is #include'd, the expedient choice has
 * been made to do the cpumask_t typedef here.
 */

typedef       __cpumask_t     cpumask_t;
#endif  /* FREEBSDV>=520 */

The build error is complaining that __cpumask_t has no storage class or
size, so it obviously can't use it in a typedef.  So I went looking for
__cpumask_t on my system and came up empty.  It's never defined
anywhere, neither in the system headers nor or the lsof source itself.
The only place it appears is in machine.h where it's used as the source
of a typedef.

root at basement:/usr/ports/sysutils/lsof/work/lsof_4.72A.freebsd$ find / -name "*.h" | xargs grep "__cpumask_t"
/usr/ports/sysutils/lsof/work/lsof_4.72A.freebsd/dialects/freebsd/machine.h:typedef __cpumask_t     cpumask_t;
/usr/ports/sysutils/lsof/work/lsof_4.72A.freebsd/machine.h:typedef __cpumask_t     cpumask_t;
root at basement:/usr/ports/sysutils/lsof/work/lsof_4.72A.freebsd$ 

My next step was to look for the system's definition of cpumask_t, with
the intent to ''fix'' machine.h.  But I also can't find that type defined
anywhere.  The funny thing is, cpumask_t isn't actually *used* in the lsof
source tree.  I commented out the typedef and now everything builds
fine.  So my question is now, where *should* __cpumask_t and/or cpumask_t get
defined, why don't I have those definitions, and why does everyone else
seem to have them?

--Mike

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-ports/attachments/20040426/6dfc99a3/attachment.bin