all the fields used by fam. It's easier to use the statfs interface for
now.
Don't explicitly link against libstdc++, it is the job of the C++
compiler to do that.
+ Use the mnttab(4) interface instead of mntent.
+ C++ syntax cleanups to appease the SunPro compiler.
+ Use MAXPATHLEN instead of NAME_MAX which can be unimplemented on strict
POSIX systems.
+ Use POSIX inttypes.h if BSD types are not available.
+ RPC includes and C++ namespace fixes.
+ Use dirent(3) instead of older sys/dir.h interface.
+ Avoid `sun' namespace collision.
Patch based on the work of Robert Lillack and others, described in
http://mail-index.netbsd.org/tech-pkg/2004/08/20/0005.html and tested with
SunPro and gcc.
directories in "real time", without having to periodically poll(2) for them
after several seconds. This improves the responsiveness of applications
using famd (specially GNOME) when changes to files occur from the "outside"
(they are notified from changes immediately), or even from the application
itself (for example, have you ever tried to rename several files from
Nautilus? just a PITA).
To enable kqueue, you have to pass the 'kqueue' option to the package.
I'm not enabling it by default because it needs testing (but it should be
enabled in a future). Furthermore, I'd like to send these patches to the
FAM developers for comments too. And a review from somebody knowing kqueue
could be good! (this is the first time I use the kqueue interface, so I'm
not sure if everything is right).
A description on how this works can be found in the files/IMonKQueue.c++
file. Note that, due to FAM's design, the easiest way to do this change
is to "emulate" imon functionality.
While here, add an rc.d script for famd, in case the user prefers to run
it at system startup instead of from inetd (I created it while developing
the kqueue functionality, so it's a good moment to add it). Adjust the
MESSAGE accordingly.
Bump PKGREVISION to 4.
of mounted filesystems in a local variable that was lost between calls to
the emulated getmntinfo function.
I'm not sure if this fixes any visible problem, but it probably should.
The code was randomly reading from memory before (running with debugging
enabled exposed the problem, for example).
Bump PKGREVISION to 3.
of /tmp! If you have been using GNOME, you may want to 'rm -f /.fam*' to
clean them up.
The problem is caused by two different issues, related to a call to tempnam:
- The code uses a HAVE_UNSETENV define to check if the unsetenv function
is available; however, the configure script does not check for it, so
unsetenv is never used. Fix the configure script to check for unsetenv.
- If unsetenv is not available, it does putenv("TMPDIR=") before calling
tempnam. The code expects that this call *unsets* the variable from the
environment (which happens in Linux), but instead it is *set* to an empty
value, causing the creation of files in the root directory. Fix this by
explicitly setting TMPDIR to /tmp.
Bump PKGREVISION to 1.
- PR pkg/23017
- PR pkg/23495
- PR pkg/23496
- PR pkg/24578
The patches for 23017 and 24578 were provided by reed@. Although rh@
suspended PR 23017 saying that pkgsrc does not support installation to
/usr, this still needs to be fixed because the same problem occurs
even when LOCALBASE != /usr.
Changes since 2.6.10:
Changed NOGROUP usage in src/Cred.c++ to -1 as NOGROUP is not defined
in IRIX.
(mraymond@sgi.com)
removed const modifier from Scheduler.h to fix GCC 3.x build.
(wardle@sgi.com)
created new GNU autotools files:
- test whether we have IMon rather than testing OS.
- test whether we have sockaddr length members rather than testing OS.
- install daemon in sbin rather than bin.
- install manual pages in typical GNU/Linux sections.
- rename daemon to famd.
- rename fam source directory to src.
- rename libfam source directory to lib.
- remove editconf and util subdirectory.
- remove rpm target and build subdirectory.
- remove support target and subdirectory.
[wardle@sgi.com] (wardle@sgi.com)
fixed warnings about typenames in BTree.h.
(wardle@sgi.com)
regardless of the result of sys/statvfs.h availability test. It
results in two definitions of HAVE_STATVFS in confdefs.h. Check the
availability of sys/statvfs.h only on IRIX platforms to ensure that
HAVE_STATVFS is defined at most once.
- removed const from Scheduler.h to fix GCC 3 build.
- use GID_NOBODY if NOGROUP is undefined to build on SysV-like operating
systems including IRIX.
Changes in the package:
- Install configuration file in place.
- Tell the user how to setup and enable FAM.
fam-2.6.9
added idle_timeout and nfs_polling_interval options to fam.conf.
[joup@bigfoot.com, wardle@sgi.com] (wardle@sgi.com)
use daemon() as an alternative to _daemonize() to background the
fam process if it is available.
[wardle@sgi.com] (wardle@sgi.com)
allow for different socket structure on FreeBSD and other small
changes to assist FreeBSD build.
[daverufino@btinternet.com, ports@geeksrus.net] (wardle@sgi.com)
fam-2.6.8
#include some standard header files to help build on FreeBSD.
(wardle@sgi.com)
improved group handling from IRIX.
(wardle@sgi.com)
untrusted user's GID is set to NOGROUP if getgrmember() gave FAM
an invalid or inconsistent GID.
(wardle@sgi.com)
should now handle strerror returning null.
(wardle@sgi.com)
initalize libtool earlier in the configure process to avoid binaries
being created with a .C extension on some systems.
[alexl@redhat.com] (wardle@sgi.com)
added libstdc++ to linker flags to make libfam always build as a
C++ library.
(wardle@sgi.com)
added CLEANFILES make target.
(wardle@sgi.com)
/etc/mtab (or equivalent) is now opened read-only.
(wardle@sgi.com)
Approved by wiz.
