Pierre Labastie wrote:
On 05/03/2016 21:31, Bruce Dubbs wrote:
Ken Moffat wrote:
On Fri, Mar 04, 2016 at 09:46:40PM -0600, Bruce Dubbs wrote:
Ken Moffat wrote:
This time, I took a look at the Makefiles (in build/), but AFAICS
only the toplevel Makefile defines install-libs, as
install-libs-recursive, and I cannot find any other references to
install-libs and therefore no explicit rule for the recursive
variant - I guess that -recursive perhaps tells make to run
install-libs in all the directories, but I cannot see any
instructions for that in any of them, so I'm no wiser about what
command is segfaulting. I'm obviously missing something. Anybody,
please ?
Have I sent you my mount-virt.sh, umount-virt.sh, and clean-mnt-lfs.sh
scripts? Are you using jhalfs or your own scripts?
I don't have them, and I'm using my own scripts whichs seem to work
well from LFS hosts¹
At the moment I'm backing up one of my rc2 systems after rebuilding
the victims of openssl-1.0.2g, and I've just run
make --debug=iv DESTDIR=/scratch/ken/E2FSPROGS2 install-libs 2>&1 |
tee libslog
I am amazed at the amount of output, but after about 2000 lines the
first part of the install seems to happen:
No implicit rule found for '../../../lib/et/com_right.c'.
File 'installdirs-elf-lib' does not exist.
Target 'installdirs-elf-lib' is double-colon and has no
prerequisites.
Must remake target 'installdirs-elf-lib'.
make[1]: Entering directory
'/scratch/ken/e2fsprogs-1.42.12/build/lib/et'
MKINSTALLDIRS /lib /usr/lib
mkdir /scratch/ken/E2FSPROGS2
mkdir /scratch/ken/E2FSPROGS2/lib
mkdir /scratch/ken/E2FSPROGS2/usr
mkdir /scratch/ken/E2FSPROGS2/usr/lib
Successfully remade target file 'installdirs-elf-lib'.
Looking for an implicit rule for
'../../../util/install-symlink.in'.
Of course, /lib and /usr/lib already existed in chroot.
and then about 400 lines later
No implicit rule found for '../../../util/install-symlink.in'.
Must remake target 'install'.
INSTALL-ELF-LIB /lib/libcom_err.so.2.1
SYMLINK /lib/libcom_err.so.2
SYMLINK /usr/lib/libcom_err.so
LDCONFIG
Successfully remade target file 'install'.
File 'install' does not exist.
Looking for an implicit rule for
'../../../lib/et/compile_et.sh.in'.
etc
Unfortunately, it looks like that part DID happen in the chroot.
AFAICS, the next hunk of activity is
Must remake target 'installdirs'.
MKINSTALLDIRS /usr/lib /usr/include/et /usr/share/et /bin
/usr/share/man/man1 /usr/share/man/man3
mkdir /scratch/ken/E2FSPROGS2/usr/include
mkdir /scratch/ken/E2FSPROGS2/usr/include/et
mkdir /scratch/ken/E2FSPROGS2/usr/share
mkdir /scratch/ken/E2FSPROGS2/usr/share/et
mkdir /scratch/ken/E2FSPROGS2/bin
mkdir /scratch/ken/E2FSPROGS2/usr/share/man
mkdir /scratch/ken/E2FSPROGS2/usr/share/man/man1
mkdir /scratch/ken/E2FSPROGS2/usr/share/man/man3
mkdir /scratch/ken/E2FSPROGS2/usr/lib/pkgconfig
Successfully remade target file 'installdirs'.
Of those directories, all except /usr/include/et and /usr/share/et
should have already existed in chroot.
And those two /et directories did NOT get created.
[1.] Full disclosure - my logging fixes for grep-2.23 did not do the
whole job, because I failed to notice that both callers piped the
output through grep. My latest changes seem correct in completed
systems, and in chroot for as far as I got, but I'm again getting
some ls errors on screen in my intochroot scripts for passwd and
groups although the final logs look acceptable.
In chroot (I've mounted it at /mnt/lfs from an LFS system to take a
look at the evidence) -
MKINSTALLDIRS = $(top_builddir)/../config/mkinstalldirs
which matches the build I am using for DESTDIR.
And config/mkinstalldirs is just a script from 1993-9, and it invokes
mkdir.
For the moment, I am none the wiser - that definitely sounds like
the sort of thing which selinux might have stopped, but I disabled
selinux and rebooted for the second attempt.
It would be interesting if you could try a jhalfs build to see it that also
has a problem at the same place.
I could get a Fedora iso and try it in a gemu session. Do you want me to do
that?
I have a VM with Fedora 21. Is it expected to give the same issue?
If so, I can test build with jhalfs.
I don't know if Fedora 21 gives the problem or not. I think I've had
students use that successfully, but Fedora 23 is the one that gave Ken a
problem.
-- Bruce
--
http://lists.linuxfromscratch.org/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
