On Tue, Jul 25, 2017 at 12:04:27PM +0200, Florian Weimer wrote: > But it's not clear if the HPC community wants to run > containers/namespaces at all. Exploring the container-less approaches for different but related purposes[1], I just did what turned out to be a practice test of installation location independence: Installing openttd as a user with an out-of-root apt and dpkg location. Steps I needed to take were roughly: * Creating appropriate sources.list, apt.conf and debconfrc files * Creating some of the basic directory structure / empty files * (repeatedly, through the next steps) run `apt install openttd` * Adapt some postinst scripts * Bend paths to log files to the user-root location and similar * Drop python bytecode precompilation and ldconfig updates * The only thing I needed to do as root: Move away /var/lib/ucf and /etc/apt/apt.conf.d/* files as I didn't find a way to override them (as they're in preinst and probably hardcoded, respectively) * Run openttd with appropriate PATH, LD_LIBRARY_PATH and (as the game itself is not installation location dependent) working directory I think that if we can, if we want to, largely support location independent installations for the next release. That would help both user-installable packages and simplify creation of any of the container packages (if I understand them correctly). It seems most of the work on infrastructure and library packages could be done in generic places like postinst snipplets, some (but well testable) would be needed in particular packages like ucf, and then applications could be user-installable (or otherwise usable for position-independent installations) as soon as they become position-independent by themselves. Best regards chrysn [1]: I'd like to use sid versions of games on stable boxes w/o always needing to go through a changeroot -- I shouldn't have written all those tank programs. -- Kevin Flynn
Attachment:
signature.asc
Description: PGP signature