CLN: Fixes to conda environment names

[datapythonista]

We have some inconsistency regarding the names of the conda environment. At the moment, all environments for the CI are named pandas, and the environment created locally following the set up instructions (and using environment.yml) is named pandas-dev.

Then, inside several scripts (e.g. ci/code_checks.sh) we do a source activate pandas. Which means that the script works well in CI jobs, but it fails locally unless we also have a pandas environment locally.

Another thing is that in the CI config we use a CONDA_ENV environment variable, and we ignore the conda environment names in the yaml files.

What I would do is:

• Use pandas-dev as the environment name in all our conda yaml files
• Create the environments with the environment name in the used yaml file (do not override with -n $CONDA_ENV as we do now)
• Do not use source activate inside the scripts (use it in the steps in the CI config files when needed). This way when running scripts locally, the current environment will be used

@jreback @TomAugspurger @jorisvandenbossche any objection?

[jreback]

this is ok

[charlesdong1991]

i will take a look!

[datapythonista]

Thanks @charlesdong1991, let me know if you need clarifications or help.

[charlesdong1991]

Hi, @datapythonista , sorry, i think i do need your help, following your explanation above, i did a small modification to change pandas to pandas-dev, not sure if it's the right place to do... and appreciate a lot if you could give more clarifications

[datapythonista]

The main changes required are:

• In ci/deps/*.yaml change the name of the environments from pandas to pandas-dev
• Do a grep for CONDA_ENV, and remove the -n $CONDA_ENV in the conda env install calls. And remove the definitions of it, as it won't be used anymore.
• Do a grep for source activate and move the activations inside the scripts (mainly in ci/) to the CI configuration files, in the place before those scripts are called

For example, in ci/incremental/build.sh there is a source activate $CONDA_ENV. This should be removed from there, so if we call the script in our machines the active environment is used. And for the CI to work, a source activate pandas-dev should be placed in .travis.yml before the call to build.sh, so the environment is active when that is called.

[h-vetinari]

I'd like to add one more location to this: doc/source/style.ipynb
For compiling the docs, I always have to change the name of the kernel in there, to make ipython use the right one:

  "metadata": {
   "kernelspec": {
    "display_name": "Python 3",
    "language": "python",
+   "name": "pandas-dev"
-   "name": "python3"
  },

[TomAugspurger]

You should be able to fix that with a jupyter kernelspec command. Kernel names don’t have to be tied to environment names, and I think python3 is the default.

…

________________________________ From: h-vetinari <[email protected]> Sent: Thursday, November 22, 2018 10:28 To: pandas-dev/pandas Cc: Tom Augspurger; Mention Subject: Re: [pandas-dev/pandas] CLN: Fixes to conda environment names (#23851) I'd like to add one more location to this: doc/source/style.ipynb For compiling the docs, I always have to change the name of the kernel in there, to make ipython use the right one: "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", + "name": "pandas-dev" - "name": "python3" }, — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#23851 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ABQHIqoDu8ezsDhx5buZmCRhMElxq13qks5uxtCegaJpZM4Yueaf>.

[h-vetinari]

@TomAugspurger Yes, of course, but I don't want my python3 kernel spec to point to the dev environment. If we're already talking environment names (and the default from environment.yml), I think this is a reasonable adaptation.