Dockerfile: Remove unneeded installation of python3-pip

[stweil]

Fixes: f1527a2 ("upgrade base image to 20.04")

[bertsky]

Duplicate of #1003

and remove unneeded installation of python3-pip.

Why?

[stweil]

Why?

Because it is unneeded as the commit message says? It was (re-)added in commit f1527a2 without giving a reason.

[stweil]

PR (and commit message) updated to fix conflict with preceding PR.

[bertsky]

Because it is unneeded as the commit message says?

It says:

Remove unneeded installation of python3-pip

That's not an explanation. Why are we investing so much time to remove a tiny package that is usually needed (except with the python -m pip workaround)?

It was (re-)added in commit f1527a2 without giving a reason.

IMO the burden for an explanation is still on your side (even if it had been removed earlier).

[bertsky]

This did create a regression – see new CI failure.

So even if I can see your argument that in Docker we want /usr/local as venv and therefore must avoid conflict with /usr – that now necessitates doing your python3 -m pip trick in in core as well.

[kba]

😆 OK, that was a long way to to go to get back to where we were before we started messing with python3-{pip,venv}. I'll reintroduce it now, so I can end the work day with a working docker image

[bertsky]

So even if I can see your argument that in Docker we want /usr/local as venv and therefore must avoid conflict with /usr – that now necessitates doing your python3 -m pip trick in in core as well.

On top of that: in ocrd/core, we don't even manage venvs – everything is native/system-wide. So at least here we do need python3-pip.

[stweil]

Why does the docker build try pip3 install --upgrade pip setuptools outside of a virtual environment?
And why did we switch from pip to pip3 again?

I am afraid both goes in the wrong direction. We should stick to using Python only in virtual environments, and then we can always use pip.

[kba]

I will investigate it further tomorrow, but no more hotfixes today :)

[bertsky]

Why does the docker build try pip3 install --upgrade pip setuptools outside of a virtual environment?

Like I said: in ocrd/core, we have no venv, only a system-wide installation. (Which is arguably better than the pseudo-venv in ocrd_all, but there we also need sub-venvs, so...)

And why did we switch from pip to pip3 again?

Because that's what python3-pip delivers. pip is not available (unless someone installs Python 2 or sets up their local venv).

[bertsky]

I will investigate it further tomorrow, but no more hotfixes today :)

No need to: the only failure now comes from core-cuda, which we already knew would fail, because we (intentionally) did not update its base image to Ubuntu 20, so OpenCV does not build anymore. – Like I said, I am still investigating whether we can actually move towards cuda-11 libcudnn8.

[stweil]

that's what python3-pip delivers. pip is not available

I just tried it in the Docker container, and pip is available:

root@a3351d49b876:/# ls -l /usr/bin/pip*
-rwxr-xr-x. 1 root root 365 Feb 28 10:41 /usr/bin/pip
-rwxr-xr-x. 1 root root 367 Feb 28 10:41 /usr/bin/pip3

[bertsky]

I just tried it in the Docker container, and pip is available:

In which container?

[stweil]

I tried ocrd/core and its base container (that one after installing python3-pip).

[bertsky]

I tried ocrd/core and its base container (that one after installing python3-pip).

Still don't know an which state of the Dockerfile. This is becoming tedious. See original bug if you want to help.

[stweil]

pip3 was needed for Ubuntu 18.04 when it was used without a virtual environment because its python3-pip did not install /usr/bin/pip. The reason is simple: old Linux versions reserved pip for Python 2. The docker build failed because it was based on an NVIDIA container which was based on Ubuntu 18.04. cd /usr/bin && sudo ln -s pip3 pip would have fixed it, too.

Ubuntu 20.04 and other current Linux distributions no longer provide Python 2. Its python3-pip installs /usr/bin/pip.

So with current supported software or when using virtual environments (my preferred way) pip3 is not needed.