Closed orbeckst closed 5 years ago
I updated the requirements for installation instructions after today's conference call.
I updated the issue after yesterday's conference call.
@dotsdl already started with INSTALL.md.
Continuing on this tomorrow evening. Thanks @orbeckst for laying out the requirements!
@richardjgowers @orbeckst are we good to proceed with Python 3.7? I don't think we currently test for this in Travis.
To do yet:
conda
packages.Although 3.7 should work in principle, I'd rather go with 3.6 which is extensively tested.
There's an issue or PR to add 3.7 to the test matrix but I forgot what the status is.
-- Oliver Beckstein email: orbeckst@gmail.com
Am Oct 23, 2018 um 23:12 schrieb David Dotson notifications@github.com:
@richardjgowers @orbeckst are we good to proceed with Python 3.7? I don't think we currently test for this in Travis.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.
conda-forge is still updating everything for python 3.7. status. It is combined with a switch to the new compiler infrastructure in conda. They are making good progress but they don't have as much CI time as they would like and a lot of packages to update.
Ah okay, we'll instruct participants to install 3.6 then. Agreed.
Added package installation instructions. Decided to use conda
for installing only MDAnalysis until we have conda packages for pmda
and MDAnalysisData
. Since the latter are pure Python we should be okay there.
Also, jupyter
and dask
come with Anaconda, so those are already in place.
Note: installing mdanalysis
with conda
will auto-downgrade the installed python:
The following packages will be DOWNGRADED:
python: 3.7.0-hc3d631a_0 --> 3.6.6-h5001a0f_0 conda-forge
@dotsdl some feedback, roughly from top to bottom on the installation instructions:
Can we say that people need Anaconda, rather than Python?
Also from my Windows 10 VM, there's an "anaconda shell" which seems to be good enough to do simple command line things from
From Windows it looks like there's a start menu shortcut to Jupyter notebook (inbetween the Candy Crush shortcuts) which might be the best way to launch this?
In which case, are we even using a bash shell at all? Most stuff is from notebooks, git interaction can be via Atom I think.. only thing I can think of is running pytest, but surely that can be via the "IDE" too..
@dotsdl @richardjgowers Windows version Anaconda has a navigator which contains Jupyter notebook. And we can do pip install
or conda install
using the Anaconda prompt("anaconda shell").
Current problem is that we don't have a windows conda installation for MDAnalysis and PMDA. It seems that Microsoft Visual C++ 14.0 is required to install MDAnalysis using pip install
. (I'm not sure whether C++ 14.0 is required if we use conda install
.)
@VOD555 Yeah I've managed to pip install a source distribution of MDAnalysis on Windows with visual c++ 14 too. Ideally we get a Windows conda build working...
@richardjgowers if you make a minor release of MDA we can publish a windows build on conda-forge.
@kain88-de yeah I just want to get (at least reading) PDB files working: https://github.com/MDAnalysis/mdanalysis/pull/2127
Pmda is now installable via conda
The data package is now also on conda forge. Everything is now conda installable.
@kain88-de thanks for PR #29 – merged.
https://github.com/MDAnalysis/WorkshopHackathon2018/issues/9#issuecomment-433640230
Can we say that people need Anaconda, rather than Python?
Yes, although we should really say that we install it for a full Python-based scientific computing software stack.
In which case, are we even using a bash shell at all? Most stuff is from notebooks, git interaction can be via Atom I think.. only thing I can think of is running pytest, but surely that can be via the "IDE" too..
I am not familiar with any of these shells and my experience is, that once something goes wrong, you just want to open a config file in vi
or run the shell commands that you know work. We don't have to spend time teaching the shell, I think, but it should be there (and git-bash makes it easy!)
We don't have to emphasize the shell but I think it will make our life easier having it. It also allows us to give people (almost) identical instructions across Windows, macOS, and Linux.
The installation instructions don't render nicely in GitHub markdown because of highlight sections. We should put them into a website. Should we use Jekyll as static site generator (possibly using the MDA theme)? Or use Sphinx (like the SPIDAL-MDAnalysis tutorial) but then this has to be recoded into reST. @dotsdl ?
EDIT: @dotsdl please open a new issue describing how you'd want to proceed
WRT the shell: the majority of attendees already have some familiarity with python, so I think it's ok to assume that most of them, being PhD students, are already (becoming) familiar with the terminal
In which case, are we even using a bash shell at all? Most stuff is from notebooks, git interaction can be via Atom I think.. only thing I can think of is running pytest, but surely that can be via the "IDE" too..
I am not familiar with any of these shells and my experience is, that once something goes wrong, you just want to open a config file in
vi
or run the shell commands that you know work. We don't have to spend time teaching the shell, I think, but it should be there (and git-bash makes it easy!)We don't have to emphasize the shell but I think it will make our life easier having it. It also allows us to give people (almost) identical instructions across Windows, macOS, and Linux.
@dotsdl please check the formatting, https://www.mdanalysis.org/WorkshopHackathon2018/INSTALL is garbled towards the end.
We need a single set of installation instructions as a web page.
The installation instructions (link) will be send as part of the info pack email.
Environment
See for example
Conda Packages
Old installation instructions
For comparison, see what we had for