Installation "warning" messages on Anaconda 5.2

[castillohair]

On Windows 10 (April 2018 Update) with Anaconda 5.2, running the installation script results in the following red "warning" (it's red and obnoxious but doesn't crash or stop anything) in the middle of all the text:

distributed 1.21.8 requires msgpack, which is not installed.

On OSX (High Sierra) with Anaconda 5.2, running the installation script produces these two red warnings:

distributed 1.21.8 requires msgpack, which is not installed.
grin 1.2.1 requires argparse>=1.1, which is not installed.

In both cases, installation proceeds normally despite these messages. The FlowCal examples run perfectly, and unit tests pass.

As a reminder, both installation scripts contain the following two lines:

python -m pip install -r requirements.txt --no-cache-dir
python setup.py install

I traced back the origin of the error to pip itself, the utility that we use to install dependencies. In both cases, running python -m pip install -r requirements.txt --no-cache-dir on a terminal produces the warnings.

pip as included in Anaconda 5.2 is in version 10.0.1 (see here and here). However the latest version is 18.0 (see here). Upgrading pip with the following command

python -m pip install --upgrade pip

Results in complete elimination of the warnings in both OSs.

The conclusions are:

• These warnings are coming from pip, and are not a problem with FlowCal. For this reason, I think the release of FlowCal can proceed without solving this.
• However, one easy thing that can be done to alleviate this is to upgrade pip in both installation scripts before installing requirements and FlowCal itself, i.e., the first line should be python -m pip install --upgrade pip.
• Lastly, pip's huge version mismatch between the one included with Anaconda and the most recent one reveals that keeping pip up to date is not a priority for the Anaconda developers. Furthermore, conda is the package manager endorse by Anaconda and not pip. While the official documentation does not prevent using pip when conda is present, user reports warn against this practice. It seems that switching installation to conda might be the most correct thing to do and might alleviate potential issues in the long run. But I haven't read enough to be sure.

To summarize, the solution to this issue could be one of the following:

• Nothing, if we decide this is not a problem because it doesn't break FlowCal.
• Upgrade pip at the beginning of both installation scripts.
• Switch to conda for installing requirements.
• Something else.

[JS3xton]

I don't like upgrading pip at the beginning of installation scripts. That sounds like a hack that will only be necessary in the short term.

Using conda where possible seems like a good idea to me.

[castillohair]

The most recent version of Anaconda as of now (2018.12) includes pip 18.1, so this is not an issue anymore. Still, the idea of switching to conda might still be worth considering in the long run. I will close this issue and open a new one dedicated just to switching to conda.