On Sun, 2011-09-04 at 20:06 -0400, Joey Hess wrote:
> I don't want to waste your time, but there are enough changes here
> that a split patchset would improve reviewing.
Perhaps, but they're really not huge.
> > if [ "$os" = "linux" ] ; then
> > - if [ ! -e "$moddir/modules.dep" ]; then
> > - echo "no $moddir/modules.dep. This file is required by kernel-wedge" >&2
> > - exit 1
> > + if [ -e "$moddir/modules.dep" ]; then
> > + modulesdep=$moddir/modules.dep
> > + else
> > + modulesdep=$tmpdir/modules.dep
> > + depmod ${SOURCEDIR:+-b $SOURCEDIR} -V $installedname -n >$modulesdep
>
> Shouldn't depmod be passed a System.map here? My experience with d-i's
> build system's use of depmod has taught me to always provide a
> System.map to it when the running kernel may differ from the kernel
> being prepared.
According to my reading of the manual page, it only wants System.map
when checking for undefined symbols (the -e option). And strace shows
that it does not attempt to open any default System.map file.
But more importantly, the -V option is *completely* bogus, so thank you
for bringing my attention back to this command.
Ben.
Attachment:
signature.asc
Description: This is a digitally signed message part