Unset DISPLAY to avoid importing ROOT when not showing plots

[jmcarcell]

BEGINRELEASENOTES

• Unset DISPLAY to avoid importing ROOT when not showing plots

ENDRELEASENOTES

In my testing unsetting the environment variable DISPLAY seems to be enough to disable showing plots. The reason to do this is that importing ROOT is slow (on my desktop it is ~0.5 seconds) so it's better if we can avoid it.

If DISPLAY is set, then batch mode is guaranteed not to be set: https://github.com/root-project/root/blob/ee9e36b3c7b9209a5ae1f1f4f7a8e4afba104c25/core/base/src/TROOT.cxx#L875

Can you please test it @Zehvogel (since you found this originally in #207)? You can copy and paste the k4run script in this PR without building anything.

I also thought about changing the option from --interactive-root to --interactive-graphics since all graphics will be disabled (I think), but it's true that most likely it's ROOT graphics what someone wants to see.

[Zehvogel]

If DISPLAY is set, then batch mode is guaranteed not to be set: https://github.com/root-project/root/blob/ee9e36b3c7b9209a5ae1f1f4f7a8e4afba104c25/core/base/src/TROOT.cxx#L875

Do we want to rely on this implementation detail of ROOT? Or is this documented as an actual guarantee somewhere?
Also, testing anything with graphics from my side will have to wait until end of next week.

[jmcarcell]

Well it's more a Linux thing since DISPLAY is set by the display managers, and it's not set if you don't have graphics (so programs have to also handle this case). In a desktop session, f you unset DISPLAY lots of things will not work, and it is used to know where to display things (you can have multiple displays). This is why I say that in that case maybe --interactive-graphics is a better option than --interactive-root since you are disabling probably all graphics (although most of the time it is ROOT anyway). So I don't think it's that much of an implementation detail, actually it is more general as we are probably stopping any graphics output.

[tmadlener]

Do we want to rely on this implementation detail of ROOT?

Also that line has last changed (according to git) 25 years ago, when ROOT went from cvs to svn, so I would consider the behavior fairly stable for now ;). We should document that k4run unsets DISPLAY by default however to avoid surprises for users.

edit: One thing that might break with this is https://github.com/key4hep/k4MarlinWrapper/blob/main/k4MarlinWrapper/examples/event_display.py . I am not entirely sure since technically glced is started before k4run in an environment where DISPLAY is still set, but I don't know whether connecting to that from the wrapped marlin processor needs DISPLAY or not.

[andresailer]

BTW: this issue in ROOT is probably relevant for the general issue: root-project/root#18744

[jmcarcell]

I'm happy with this as it is. Once root-project/root#18744 is merged we can also use that but in the meanwhile this works as intended for me.

[jmcarcell]

Unfortunately this seems not to be so easy. Running the ported Conformal Tracking prints what seems to be harmless errors:

TUnixSystem::Se...  ERROR Can't figure out DISPLAY, set it manually
In case you run a remote ssh session, restart your ssh session with:
=========>  ssh -Y
TUnixSystem::Se...  ERROR Can't figure out DISPLAY, set it manually
In case you run a remote ssh session, restart your ssh session with:
=========>  ssh -Y

This doesn't happen when running other algorithms that create ROOT histograms and write them to a file. Also, it doesn't happen when running the CLD reconstruction. I'm not sure why it happens for Conformal Tracking, maybe something that is done related to ROOT from DDKaltest.