reassign -1 debootstrap,usrmerge

On 2019-04-09 16:53, Aurelien Jarno wrote:
> On 2019-04-09 12:27, Andreas Beckmann wrote:
> > Package: libc6-x32,libc6-i386
> > Version: 2.28-8
> > Severity: serious
> > User:
> > Usertags: piuparts
> > 
> > Hi,
> > 
> > during a test with piuparts in a --merged-usr environment I noticed that
> > installing, removing, and installing again a package shipping /lib32,
> > /libx32 will actually unmerge that directory.
> > The package will take ownership of the preexisting symlinks
> > /lib{32,x32} -> /usr/lib{32,x32} that were created by debootstrap,
> > remove them and create plain /usr/lib{32,x32} directories in the next
> > installation.
> > (/lib64 should be mostly safe due to /lib64/, but
> > perhaps on !x86 architectures)

/lib64 is an issue for at least libc6:amd64 on an i386 system. And there
are many more cases if you also consider other official architectures
and ports architectures.

> Hmm the only fault of the libc6-i386 and libc6-x32 packages (plus I
> guess all the other bi/tri-arch ones on other architectures) is to be
> the last user of those directory when being removed. They do not do
> anything tricky in their directories.
> > The preinst scripts could check whether the package is being installed
> > in a --merged-usr environment and create (dangling) symlinks if
> > /usr/lib{32,x32} is missing. And postrm remove could recreate them if
> > they went missing.

The glibc maintainer scripts are already cluttered with many ugly and
fragile workarounds to handle the co-installability of foreign and biarch
glibc. I do not want to adds more workarounds that might will make the
whole things a nightmare.

> This looks like an ugly workaround to me, and might not work if a
> package start adding files there without depending on libc6. This looks
> to me like a flaw in the usrmerge design. The base-files package is
> designed to prevent directories or symlinks to be removed, so I wonder
> if we need a usrmerge version of it.

As explained it's not a bug of the glibc package, but a design flaw of
usrmerge. I am therefore reassigning the bug to debootstrap + usrmerge.

I am not opposed to a workaround in the glibc package, kept for one
release cycle only, *once a real solution* is implemented for this bug.

Aurelien Jarno                          GPG: 4096R/1DDD8C9B       

Attachment: signature.asc
Description: PGP signature

Reply via email to