Hi Andreas,
> You shouldn't have /usr/lib/i386-linux-gnu/libGL.so.1 at this point ...
> Where does it come from? Did you try to "fix" some things manually?
No, I just de-selected all the relevant packages in aptitude, and then
pressed g ;-)
> Is it a file or symlink?
> ls -la, md5sum, timestamp, link target?
Sorry, since I need to use this computer I ended up doing
dpkg --force-depends --purge libgl1:amd64 libgl1:i386
dpkg --purge glx-diversions
apt install libgl1:amd64 libgl1:i386
> Any ideas how you got into this configuration? Didn't encounter
> something like this in my tests.
It is a bit special configuration: It is a laptop with Intel GPU, which
is synced to my main computer having an nvidia gpu. I install TensorFlow
with GPU support on the nvidia/main machine. I sync the stuff in
/usr/local with unison.
I want to be able to run tensorflow on my laptop, too, but with the
installation from the nvidia machine it does not work due to missing
libraries. So installed the missing libraries from nvidia, but use the
intel driver.
Now, on the main desktop I can run tensorflow with GPU support, and on
the laptop I run the same, but due to missing nvidia card it falls back
to using CPU only.
I admit this is a special setup, but apt allowed me to install the
nvidia libraries in parallel to the intel driver and mesa gl stuff.
I tried to revert the whole bunch due to strange problems on my desktop
(Cinnamon) which does not properly show notifications, some programs
don't start with very strange exception related to X. So I thought that
the nvidia libs might be a problem and reverted ...
Happy and peaceful holidays, and a great start into the new year!
Norbert
--
PREINING Norbert http://www.preining.info
Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13
