Teaching basic lab skills
for research computing

Fixing 14 Repositories

We cloned our workshop repository at the beginning of the workshop at Spelman to give everyone an identical directory tree and contents to explore, and to distribute the sample data files. On the first day of our workshop, about a third of the learners (all running Windows) could not clone the workshop repository: git clone was failing for them with the error:

$ git clone http://github.com/wltrimbl/2014-06-10-spelman
fatal: cannot create directory at 'misc/aux': Invalid argument
warning: clone succeeded, but checkout failed.
You can inspect what was checked out with 'git status'
and retry the checkout with 'git checkout -f HEAD'

I had two problems:

  1. I couldn't reproduce the problem because I only had a Mac with me.
  2. I wanted to make sure other workshops did not have the same problem.

1. A virtual crutch for the Windows-challenged: EC2

Amazon rents computers for between $0.02 / hour to $2 / hour, and for $4 I could rent a Windows computer for two days.

Starting a Windows machine was a matter of giving Amazon my credit card, creating an SSH keypair, starting one of the Windows Server 2008 images, and getting the EC2 dashboard to tell me the Administrator Password for the machine I'd rented.

Per the instructions, Microsoft Remote Desktop (a free app) lets you log in and use the desktop of a remote Windows machine. It behaves much like virtual box, except that the vitual box is in a data center in Virginia and not on my laptop.

This gave me access to a bare-bones locked-down Windows machine. I had to change the security settings to allow downloads from the browser, and this let me install git bash. When I tried git clone I saw the same error message and got the same partial checkout that my students had.

It turns out this was a known issue that was already fixed with PR#464 and PR#465, but I had forked this workshop's repository before these fixes were applied, so I found a workaround: correcting the problem in the upstream repository, then instructing everyone to run:

$ git fetch
$ git checkout -f gh-pages

The workaround didn't repair the repository—git was still unhappy—but it did syncrhonize all the files for the rest of the class.

Bottom line: instructors who have never laid hands on a Windows machine can, for a modest fee, test all the install and configuration quirks on those versions of windows offered by Amazon EC2.

Next I wondered: if I forked the repository when it had a problem, how many other instructors did the same?

2. How many workshops suffer from the same issue?

Once I knew what the problem was I realized I could test repositories by cloning them on a Linux machine and testing each to see if it had the aux directory. cloneallthebootcamps.sh does this, though not very intelligently. It clones all the future workshop repos (on 2014-06-12, 27 repositories, requiring requiring 1.6Gb on disk) and checks for aux in the default branch.

Initially, 14 were affected. Two days later, that number is down to 5:

$ find 2014* -name aux -type d

Out of curiosity (or mischief), and empowered by a script that would download dozens of repositories for little effort, I ran the tests on old bootcamps. Looking at 55 repositories in 2014 before 2014-06-06 the only ones I find with the evil aux directory are:


This means that we identified the flaw in the middle of the third workshop, and had it corrected by the fifth.

What did we learn?

  1. "I don't have a windows machine" is not really a credible excuse for not testing (and, more importantly, becoming familiar with the install and invocation procedure) on Windows. "I don't want to do it because it takes an hour to initialize and install and it's an unfamiliar computing environment" is now the more honest response.
  2. Serious problems can propagate into many repositories before being detected because of the lag between setting up the website (typically weeks in advance) and that of the workshop.
  3. It's a little bit of a pain to fix the same problem in 14 repositories.

Editor: and #4: a few minutes of scripting can save hours of pain, and not just for the person writing the script. Many thanks to Will for looking for this in everyone else's repositories as well as his own.

Dialogue & Discussion

Comments must follow to our Code of Conduct.