glib2.0:armel uninstallable on systems that don't support executing armel code
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
glib2.0 (Debian) |
Fix Released
|
Unknown
|
|||
glib2.0 (Ubuntu) |
Fix Released
|
Low
|
Martin Pitt |
Bug Description
glib-2.0 declares itself Multi-Arch: same, i.e co-installable. But in fact this does not work because it cannot be installed on anything other than the native architecture.
Amongst other things this means that no package build-depending on glib2.0 can be built (becuase the build-deps don't install).
$apt-get install libglib2.0-0:armel
Setting up libglib2.0-0:armel (2.31.20-0ubuntu2) ...
/var/lib/
/var/lib/
dpkg: error processing libglib2.0-0:armel (--configure):
subprocess installed post-installation script returned error exit status 127
The problem is that the postinst contains:
case $trigger in
# This is triggered everytime an application installs a
# GSettings schema
;;
# This is triggered everytime an application installs a GIO
# module into /usr/lib/
# backwards-
/usr/lib/
/usr/lib/
are both ELF executables and thus not executable on the machine you are cross-installing onto.
If the output of these commands is arch-independent then a good fix would be to change these lines to be:
/usr/lib/
/usr/lib/
so that the version for the arch being installed-onto always gets run.
However if it is not arch-independent then the postinsts should arrange to only run these commands when the package is being natively-installed. If the output for all the versions installed needs to be present then we need to arrange some way of making this work.
It may be that moving these binaries out into the libglib2.0-bin package is the right thing to do?
description: | updated |
summary: |
- glib2.0:armel uninstallable on other architectures + glib2.0:armel uninstallable on non-native architectures |
Changed in glib2.0 (Debian): | |
status: | Unknown → Fix Released |
Changed in glib2.0 (Ubuntu): | |
assignee: | nobody → Martin Pitt (pitti) |
It can be installed on lots of things that aren't the native architecture, the only requirement is that you're able to run the code - which is straightforward to accomplish using qemu-user-static, isn't it?