Update models docs to core 2.22.0

[kba]

No description provided.

[bertsky]

Should we explicitly mention presets as a special kind of resource?

And should we give a perspective on practical resource management with ocrd_all native and ocrd/all containers here? (Just asking, really.)

[EEngl52]

looks good to me. We should also adjust the example calls in the workflow guide

[kba]

Should we explicitly mention presets as a special kind of resource?

Good point, I haven't been considering parameter presets, I'll have to make sure that it is properly covered and then document it. So I'll make another pre-release and do a proper release once I am sure we have this covered.

And should we give a perspective on practical resource management with ocrd_all native and ocrd/all containers here? (Just asking, really.)

Yes! One reason for resmgr was to make it easier to mount models into docker and we should document that this is now possible.

[kba]

One reason for resmgr was to make it easier to mount models into docker and we should document that this is now possible.

d886e68 @bertsky I don't have docker available right now, so I haven't tested this. Can you check please whether this works? In particular, what is $HOME in ocrd_all? Since we advise to use the docker run --user flag, I am fairly certain, it should be / but not entirely sure. Thanks!

[bertsky]

d886e68

Forgot to push?

I don't have docker available right now, so I haven't tested this. Can you check please whether this works? In particular, what is $HOME in ocrd_all?

HOME expands to /root in ocrd/all.

Since we advise to use the docker run --user flag, I am fairly certain, it should be / but not entirely sure. Thanks!

Yes, with -u HOME becomes /.

[kba]

Forgot to push?

Yup, sry 😊 pushed it now.

Yes, with -u HOME becomes /.

Thanks for checking.

[bertsky]

There were various inaccuracies in the commands, but more importantly, I don't think we should actually recommend that 2-step approach. You would still have to natively install core independent of Docker, which defeats the use-case for virtualization in the first place. (And in the future, we even want to aggregate the database from submodules on the fly, so a full-blown native ocrd_all installation would be necessary to download resources on the host.)

Instead, this alternative formulation just explains how to mount a persistent host directory for the models, and use Docker to download models into it whenever needed.

[kba]

Instead, this alternative formulation just explains how to mount a persistent host directory for the models, and use Docker to download models into it whenever needed.

You're right, I didn't consider how you'd need a full deployment on the host to download the models outside of docker.

And IIRC for that very reason we also wanted to make --location virtualenv the default in the Docker build...

This works on the top-level venv but what about the sub-venvs? They have their own $VIRTUAL_ENV and hence the ocrd resmgr in the sub-venvs will not see the model installed to top-level $VIRTUAL_ENV/share/ocrd-resources :( Correct me if I'm wrong but I think we do need either XDG or re-introduce an absolute path like /usr/share/local/ocrd-resources ...

[bertsky]

Sorry, I was not aware the markdown rendering is different on ocr-d.de. It did not take my four-space prefix as command syntax, we must change to code blocks (inside list items!) again...

[kba]

I was afraid so but didn't check. I'll prepare a fix.

[bertsky]

Also, we still need to explain that now processors will automatically download resources they need. (This is a bit surprising given all the ocrd resmgr facilities. So it should be made clear.)

[kba]

Don't we mention "downloaded on-demand"?

[bertsky]

Don't we mention "downloaded on-demand"?

Now that you say it: yes, once, in a comment of a command for one processors' resources 😄

[kba]

Now that you say it: yes, once, in a comment of a command for one processors' resources 😄

that's all a true hacker needs : Jest aside, I added a section on that in #197