Am 29.04.2016 um 16:00 schrieb Peter Maydell: > On 29 April 2016 at 14:56, Stefan Weil <s...@weilnetz.de> wrote: >> Am 29.04.2016 um 15:54 schrieb Peter Maydell: >> >>> This means we'll build with a HAVE_FSXATTR define set, but >>> nothing in the tree tries to use that as far as I can tell: >>> "git grep HAVE_FSXATTR" returns no matches. What am I missing? >> It's used by the system headers: >> >> /usr/include/xfs/xfs_fs.h:#ifndef HAVE_FSXATTR > ...so this is a bug in the system headers that we're working > around? It would probably be useful to say so in a comment > in configure, otherwise it's liable to get ripped out in > future when somebody notices it's not used any more. > (For instance HAVE_IFADDRS_H isn't used and looks like dead > code in configure.) > > thanks > -- PMM
Is it a bug of the system headers? Or simply a design which requires users to be careful when including certain header files? Both /usr/include/xfs/xfs_fs.h and /usr/include/linux/fs.h define the same struct fsxattr, and both definitions are identical. Updating to xfslib-dev 4.3.0 did not help for Debian. This means that even with a consistent installation of Debian Testing QEMU fails to build as soon as CONFIG_XFS is defined. Of course a good comment would be helpful here, e. g. # Avoid redefinition of struct fsxattr in xfs/xfs_fs.h. # It is already defined in linux/fs.h. Stefan