Current breakage of install due to not specifying dm_control version (or versions for any dependency)

[peteflorence]

Hi folks, a few days ago, this worked, given a working underneath mujoco install:

pip install \
  mujoco_py==2.0.2.5 \
  git+https://github.com/rail-berkeley/d4rl@master#egg=d4rl

Although this repo's master branch hasn't been updated since July, the underlying dm_control dependency version is not specified, and it has since updated to mujoco 2.1: google-deepmind/dm_control@420e237

(which doesn't exist publicly yet: https://www.roboti.us/download.html)

This causes the following error when trying to run install d4rl:

Successfully built mujoco-py d4rl mjrl
Failed to build dm-control
Installing collected packages: mujoco-py, mjrl, dm-control, click, d4rl
    Running setup.py install for dm-control ... error
    ERROR: Command errored out with exit status 1:
     command: /usr/bin/python3 -u -c 'import io, os, sys, setuptools, tokenize; sys.argv[0] = '"'"'/tmp/pip-install-p1_maoog/dm-control_a87dbf57bef14d5bb352aa0e2a5bcad6/setup.py'"'"'; __file__='"'"'/tmp/pip-install-p1_maoog/dm-control_a87dbf57bef14d5bb352aa0e2a5bcad6/setup.py'"'"';f = getattr(tokenize, '"'"'open'"'"', open)(__file__) if os.path.exists(__file__) else io.StringIO('"'"'from setuptools import setup; setup()'"'"');code = f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' install --record /tmp/pip-record-gr13f_14/install-record.txt --single-version-externally-managed --user --prefix= --compile --install-headers /home/peteflorence/.local/include/python3.8/dm-control
         cwd: /tmp/pip-install-p1_maoog/dm-control_a87dbf57bef14d5bb352aa0e2a5bcad6/
    Complete output (2 lines):
    running install
    error: Header file '/home/peteflorence/.mujoco/mujoco210/include/mjdata.h' does not exist.

Could someone add versions to all dependencies here? https://github.com/rail-berkeley/d4rl/blob/master/setup.py

Also, to clarify: although one could wait until mujoco 2.1 is available, it still seems that versioning the benchmark is required, or some other mechanism for not ensuring regressions?

[peteflorence]

I made a pull request which verions dm_control to a specific hash: #126

But:

1. Probably all dependencies should be versioned?
2. I'm not sure if that specific hash should be the one?

[peteflorence]

Ah, now I see that they changed the website: https://mujoco.org/download

[Harimus]

had same issue but adding 210 just caused me this issue in dm_control
google-deepmind/dm_control#210

Edit: as a temporary workaround, setting dm-control==0.0.364896371in setup.py in d4rl worked for me.
Edit2: Probably better with dm-control==0.0.425341097

[saran-t]

Can I ask why d4rl is depending on dm_control via git rather than PyPI?

[tfboyd]

I decided to change this comment. I am a bit frustrated; but it is what it is and i knew it was not a good idea to create a dependency (even just fro an experimental test for CQL-SAC) on a github repo vs. at least a published .whl. But it is also a pain to maintain stuff so I totally understand, although being transparent in the README.md seems like a small courtesy I personally would do to set some expectations.

[ethanluoyc]

I also just want to add an extra piece of info here. Installing dm_control from git can produce weird error messages on a fresh virtual environment due to some build-time dependencies not being met. I created a PR on the dm-control side to silence this. google-deepmind/dm_control#215.