DOC: Improve compiler installation instructions

[lucyleeow]

• further improves remove confusing instructions and link. #27717 - Currently the 'Contributing to pandas', 'Installing a C compiler' section provides links to pages that DO NOT give information about how to install a C compiler.

Note:

• For Linux, we have recommended to install a group of packages, along with GCC:
  sudo apt-get install build-essential and yum groupinstall "Development Tools" whereas in the sklearn installation page have recommended to install GCC with sudo apt-get install build-essential but to install individually for Fedora etc: `sudo yum -y install gcc gcc-c++``

Which would be better?

[WillAyd]

Thanks for the PR. I'm indifferent as to whether or not some of this should be included. Good to have but main concern will be keeping relevant (ex: Xcode 4.2 mentioned was released 8 years ago)

Just out of curiosity have you tried using a conda installed compiler? Wondering if that wouldn't make configuration and installation easier for newcomers:

https://www.anaconda.com/utilizing-the-new-compilers-in-anaconda-distribution-5/

[lucyleeow]

Main concern was that the links in the current documentation direct to pages that do not actually talk about how to install a C compiler. This is confusing and I think this is a barrier for new contributors.

Happy to change doc to just link to conda and state that there are alternative ways of installing compilers and they can use if they wish?

[WillAyd]

Maybe - haven't tried personally so don't know of any limitations that would have w.r.t. pandas. Interested in trying it out on your end?

[lucyleeow]

I'm not sure if a conda installed compiler is a good option as it is designed for building conda packages. However, installing a compiler from conda forge may be a good option for Mac. I believe @ogrisel and @jeremiedbb are looking into compiler installation for Mac and then updating the sklearn contributing docs. I think I will link/copy theirs once they are done.

Meanwhile I might copy the windows section from sklearn instead of just linking to it.

[jorisvandenbossche]

@WillAyd the link about the conda compilers you give is a link about using those compilers when building packages, but not about installing those in a user (contributor) environment. However, conda-forge wrapped those in a meta-package that can be installed: https://anaconda.org/conda-forge/compilers.

But, we first need to have more testing of this I think to be sure to recommend this. In the meantime, this PR is already a big improvement over the current useless links (except for the mac part which is incorrect, maybe we should just leave that out for now).

[jorisvandenbossche]

For Mac, the https://devguide.python.org/setup/#macos link actually has some explanation about installing XCode, so maybe for this one we can keep the link until we have better instructions.

[lucyleeow]

I have replaced my Mac instructions with just the link above.

I've also changed formatting - happy to change back as well @jorisvandenbossche

[jorisvandenbossche]

Formatting looks good!

[jorisvandenbossche]

This can be removed here, as we first still need to install all dependencies which is only mentioned below

[lucyleeow]

Oh yes, that was silly of me.

[jorisvandenbossche]

I think this step is not needed for pandas, as we use setuptools and not distutils as scikit-learn does (based on https://wiki.python.org/moin/WindowsCompilers#Distutils_notes, at least the env variables is not needed for setuptools)

[jorisvandenbossche]

cc @chris-b1

[WillAyd]

lgtm @jorisvandenbossche

[jorisvandenbossche]

Thanks @lucyleeow !

We can always further try to improve it (eg if we know the conda-forge compilers can help), but merging this as it is a good improvement over the current links.