Merge from trunk@493 to unbreak linux rpm spec. Seemed a worthwhile thing to do for a release, even if Linux is a minor platform used only when *BSD doesn't have the right drivers for your hardware.

Patch from Per Thomsen to skip kqueue check on NetBSD, required because the NetBSD kqueue is not compatible with the OpenBSD kqueue. This is a temporary workaround to be revisited for 0.11.

Patch from Per Thomsen to detect statvfs. This is used by recent versions of NetBSD in preference to statfs.

Disable memory testing on Darwin -- order of destructors of static variables is nasty so debug processes segfault on termination

- Added missing Visual Studio 2003 project file

* 2003/boxbackup.suo
* 2003/win32test.vcproj
* 2003/common.vcproj
* 2003/bbackupctl.vcproj
* 2003/bbackupd.vcproj
- Updated Visual Studio 2003 project files, thanks to "Mark" 
  <boxbackup@hostingbrothers.nl>

On XFS the file permissions appear to apply also to the extended attribute data. ie. If a file is not readable then nor is the extended attribute data - even if there isn't any this is still an error.

This is not an ideal fix for this problem. Really we should notify a read-error as we do for file read permission problems and continue, but the BackupClientFileAttributes code is called from multiple locations and it is not clear that it could always report the error in such a way, nor even continue. This alternative is better than aborting the whole run. In the case where a file isn't readable this will be found and reported elsewhere anyway.

Zero any unused fields in struct stat, to avoid spurious checksum mismatches that waste bandwidth and slow down backups (committing on behalf of chris)

Bandwidth usage logging patch from Pascal Lalonde <plalonde@overnet.qc.ca>, with minor change to exception handling

Remove ppc hack and configure option. Replace dubious casts with clearer use of memcpy to make this work on all platforms all of the time.
Note that there isn't really a call to memcpy, the compiler inlines it as 2 int copies which is an insignificant overhead (even true with -O0).

Added workaround option for GCC on the PPC platform. This is a manually invoked option because I do not know under what circumstances (eg. versions etc) the problem occurs.
If the raidfile test fails in release mode but passes in debug mode then try this option and see if it fixes it.

Workaround for autoconf's unexpanded directory substitutions vs. the current Makefile not being able to expand them itself.

NOTE: Needs testing on other platforms, especially BSD.

NOTE: Can someone test that this doesn't break Windows please?

Now debug builds are once again unoptimised.

- Omit Windows-specific files on non-Windows platforms

- Use the target platform's name, rather than the build platform, in the
  generated package file names (useful for cross-compiling)

* lib/win32/emu.cpp
- Don't log a warning when statting a file that doesn't exist, as 
  bbackupquery does this to verify that the target does not exist before 
  restoring

* lib/backupclient/BackupClientFileAttributes.cpp
- Fixed a missing endian conversion that was causing file timestamps to 
  appear to be invalid, generating warnings about every file on the server,
  on Win32

* parcels.txt
- Include a couple of DLL files in the package on MinGW, which are needed
  to run the compiled binaries

* bin/bbackupquery/BackupQueries.cpp
- Fixed usage text for the "get" command

* bin/bbstored/backupprotocol.txt
* bin/bbstored/BackupCommands.cpp
* bin/bbackupd/BackupClientContext.cpp
* bin/bbackupd/BackupClientContext.h
* bin/bbackupd/BackupClientDirectoryRecord.cpp
* bin/bbackupd/BackupDaemon.cpp
* lib/backupclient/BackupStoreFileDiff.cpp
* lib/backupclient/BackupDaemonConfigVerify.cpp
* lib/backupclient/BackupStoreFile.h
* test/backupstore/testbackupstore.cpp
* test/backupstorepatch/testbackupstorepatch.cpp
- Applied changes from chris/diff-timeout-and-ssl-keepalive

* test/backupdiff/testbackupdiff.cpp
- Fixed test to match new prototype for EncodeFileDiff

Now intercept tests are never disabled.

- A very unlikely failure where getxattr returns an error code the first time
  but succeeds the second time could cause buffer overrun and corruption.

Affecting the tests only:

- On Linux (and presumably Irix) normal users can only modify attributes in
  the 'user.' namespace.
- Some filesystems can have strict limits on value size (eg. 1024 bytes on
  ext3 depending on block size)
- ENOATTR is defined in xattr/xattr.h. This file is an optional install and is
  not needed for the rest of box to use extended attributes. Provide it
  ourselves if not present.

Change handling of static buffer in filename encoding and decoding to avoid double free on exit under Darwin -- maybe a runtime issue?