(re-adding list as this certainly does not contain sensitive
information - others might add other opinions and hints, as well as my
answer should be available for everyone to find it.)
Am Mo., 22. Feb. 2021 um 14:35 Uhr schrieb <email@example.com>:
On Mon, Feb 22, 2021 at 04:57 AM, Josef Holzmayr wrote:
Whats the reasoning behind this? If its meant to be a work-around for
"my custom software totally wants it in that location", then you're
probably better off fixing your custom software to stick to canonical
paths. If its about partitioning schemes, other techniques might
apply. If its about being able to upgrade/modify python independently
from the system, then you probably want some root-in-root or container
build. But randomly picing a complex package that has system-wide
implications and saying "I want it here, screw the FHS" is not a good
I am aware that what I am asking for is a bit ugly.
The reason is that I have a small amount of memory at my disposal. I'm working with a setup with two partitions, a root-fs and an overlayed application file system. None of them has enough space for python.
Therefore I want to install it on the eMMC, which has plenty of space.
So instead of /usr which is on the root/app file system, I would install it under /media/<somewhere> on the mounted eMMC.
But maybe there exists a more elegant solution?
I personally would probably go with a build-in-build, and put some
form of application rootfs on the emmc - this could either be a simple
chroot or some more advanced form of container. This avoids nasty
breakages and update problems when the filesystems go out of version
sync. Other techniques might also apply depending on your software
rollout process, like an addtional overlay fs, or a pivot-root with
initrd, or.... it depends. But ripping out random packages and
rearranging them at random locations certainly isn't a good idea. It
already hurts when I think of the mount-and-deploy magic one would
need for this to roll out in production.