Hi,
Quoting Aurelien Jarno (2020-03-21 00:00:18)
> On 2020-03-20 22:57, Johannes 'josch' Schauer wrote:
> > would it be possible to make the libc6 preinst maintainer script
> > compatible with busybox? Currently the preinst script calls "readlink
> > -m" which is not supported by busybox. Hence the following error will be
> > thrown:
> >
> > BusyBox v1.30.1 (Debian 1:1.30.1-4) multi-call binary.
> >
> > Usage: readlink [-fnv] FILE
> >
> > Display the value of a symlink
> >
> > -f Canonicalize by following all symlinks
> > -n Don't add newline
> > -v Verbose
> >
> > I tried to prepare a patch for the preinst script but ran into a FTBFS:
> >
> > x86_64-linux-gnu-gcc-9 -shared -static-libgcc -Wl,-O1 -Wl,-z,defs
> > -Wl,-dynamic-linker=/lib64/ld-linux-x86-64.so.2
> > -B/<>/build-tree/amd64-libc/csu/
> > -Wl,--version-script=/<>/build-tree/amd64-libc/libnss_files.map
> > -Wl,-soname=libnss_files.so.2 -Wl,-z,combreloc -Wl,-z,relro
> > -Wl,--hash-style=both -L/<>/build-tree/amd64-libc
> > -L/<>/build-tree/amd64-libc/math
> > -L/<>/build-tree/amd64-libc/elf
> > -L/<>/build-tree/amd64-libc/dlfcn
> > -L/<>/build-tree/amd64-libc/nss
> > -L/<>/build-tree/amd64-libc/nis
> > -L/<>/build-tree/amd64-libc/rt
> > -L/<>/build-tree/amd64-libc/resolv
> > -L/<>/build-tree/amd64-libc/mathvec
> > -L/<>/build-tree/amd64-libc/support
> > -L/<>/build-tree/amd64-libc/nptl
> > -Wl,-rpath-link=/<>/build-tree/amd64-libc:/<>/build-tree/amd64-libc/math:/<>/build-tree/amd64-libc/elf:/<>/build-tree/amd64-libc/dlfcn:/<>/build-tree/amd64-libc/nss:/<>/build-tree/amd64-libc/nis:/<>/build-tree/amd64-libc/rt:/<>/build-tree/amd64-libc/resolv:/<>/build-tree/amd64-libc/mathvec:/<>/build-tree/amd64-libc/support:/<>/build-tree/amd64-libc/nptl
> > -o /<>/build-tree/amd64-libc/nss/libnss_files.so
> > /<>/build-tree/amd64-libc/csu/abi-note.o -Wl,--whole-archive
> > /<>/build-tree/amd64-libc/nss/libnss_files_pic.a
> > -Wl,--no-whole-archive -Wl,--start-group
> > /<>/build-tree/amd64-libc/linkobj/libc.so
> > /<>/build-tree/amd64-libc/libc_nonshared.a -Wl,--as-needed
> > /<>/build-tree/amd64-libc/elf/ld.so -Wl,--no-as-needed
> > -Wl,--end-group
> > /usr/bin/ld:
> > /usr/lib/gcc/x86_64-linux-gnu/9/../../../x86_64-linux-gnu/libselinux.so:
> > undefined reference to `gettid@GLIBC_2.30'
> > collect2: error: ld returned 1 exit status
>
> Strange. How did you try to build it?
It turned out to be a problem on my side. Sorry for the false alarm.
> > Thus, I'm now reporting this wishlist bug here before further working on
> > a fix.
> >
> > Would you be willing to accept a change that makes the preinst script of
> > libc6 compatible with readlink from busybox?
>
> On the principle yes, but it means we need to have an equivalent to
> readlink -m. Do you have a way for doing that in busybox?
Indeed I have. There exists a version for bash with an extensive test suite:
https://github.com/bashup/realpaths I ported that one to POSIX shell while
keeping the test suite and comparing it with "realpath -m":
https://gitlab.mister-muffin.de/josch/realpath
The preinst script should probably continue using coreutils readlink when it
exists and only fall back to the re-implementation in POSIX shell if the
version of readlink on the system does not provide the -m option (as it is the
case with busybox).
Since I now was able to successfully rebuild glibc, I can confirm that this is
the last puzzlepiece needed to allow to create and configure a system
containing only the following packages (and their Depends) without errors:
base-files, base-passwd, busybox, debianutils, dpkg, libc-bin, mawk, tar
So it would be great if this could be solved somehow. What do you think? :)
Thanks!
cheers, josch
signature.asc
Description: signature