The SIRO Problem

Garbage in Garbage Out

Sensible In, Reasonable Out.

PhDWare

• Don't look if anyone's done it before
• Code till it works
• Generate a figure
• Throw it away

Labware

• Understood by one genius
• Implements great science, now
• FORTRAN in any language
• Code not engineered for readability
• Can't add new science

HPCWare

• Get a 5% improvement in performance
• On a particular architecture
• Publish a scaling graph
• Selection against:
  • Readability
  • Maintainability
  • Adaptability

ConsultantWare

• Little understanding of the science
• Overengineered
• Unmaintainable by the research group

Software Carpentry

• Founded by Greg Wilson
• Volunteer organization
  • Alfred P. Sloan Foundation
  • Mozilla Foundation.
• Provides online training materials on which UCL course is based.

Software Carpentry

• Intensive "bootcamp"
• Teaches the tools needed to do computational science the Right Way
  • Readable
  • Reliable
  • Repeatable

What Version Control is For

• Managing Code Inventory
  • "When did I introduce this bug"?
  • Undoing Mistakes
• Working with other programmers
  • "How can I merge my work with Jim's"

What is version control?

Do some programming

my_vcs commit

Program some more

Realise mistake

my_vcs rollback

Mistake is undone

What is version control? (Team version)

Version Control Examples

• Working alone with git
• Publishing with git
• Teamworking in git
• Teamworking in git with conflicts

More than great backup

• Logging
• Collaboration
• Labels for code versions
• Branches for production vs play

How Things Should Be

fetch_dataset 53b6
run_model dataset_53b6
Examine_results results_28_02_13_1_53b6_98d2
archive_results latest
create_graphs results_28_02_13_1_53b6_98d2

Program or Be Programmed

• Repetition leads to Boredom
• Boredom leads to Horrifying Mistakes
• Horrifying Mistakes lead to God-I-Wish-I-Was-Still-Bored

Automation for reproducibility

• More than saving time
• Provide a rigorous description of what was done

Python

IPython Notebooks

This talk

This talk

This talk's SConscript

pandoc_slides=Builder(action='pandoc -t revealjs -s -V theme=night'+
    ' --css=night.css'+
    ' --css=slidetheme.css'+
    ' --mathjax '+
    ' -V revealjs-url=http://lab.hakim.se/reveal-js/'+
    ' $SOURCES -o $TARGET')

...
dot_figure_builder=Builder(action='dot -Tpng $SOURCE -o $TARGET',
    suffix='.png',
    src_suffix='.dot')

This talk's puppet manifest

# Checkout the repository from GitHub
  vcsrepo {"repo-$title":
      path => "/opt/gitrepos/$checkout_location",
      ensure => 'latest',
      owner => 'apache',
      user => 'apache',
      identity => '/home/ccsprsd/.ssh/id_rsa',
      group => 'apache',
      provider => 'git',
      require => [ Package["git"], File["$service_home/.ssh/id_rsa"], ],
      source => "$source",
      revision => $branch,
    }

#Run Pandoc via scons to produce the Reveal.js presentations

    exec { "build-$title":
      command => $command,
      cwd => "/opt/gitrepos/$checkout_location/$workdir",
      require => [Exec["install-pandoc"],
        Package["scons"],
        Vcsrepo["repo-$title"]]
    }

# Copy the C++ course directory into the web folder
    file {"install-$title":
      path => "$install/$target",
      source => "/opt/gitrepos/$checkout_location/$sitedir",
      recurse => true,
      require => Exec["build-$title"],
      notify => Service["httpd"],
      owner => 'ccsprsd',
      group => 'ccspx0',
    }

You don't need to test if:

• Your programs always work correctly, or
• You don't care if they're correct or not, so long as their output looks plausible, and
• You like being inefficient: the more you invest in quality, the less total time it takes to build working software

Levels of Testing

• Regression testing: Does my code work the same as yesterday
• Functional testing: Does my code match an analytic solution
• Unit testing: Does this subroutine work as expected

Test-driven development

Know what you want:

def test_range_overlap():
    assert range_overlap([ (0.0, 1.0), (5.0, 6.0) ]) == None
    assert range_overlap([ (0.0, 1.0) ]) == (0.0, 1.0)
    assert range_overlap([ (2.0, 3.0), (2.0, 4.0) ]) == (2.0, 3.0)
    assert range_overlap([ (0.0, 1.0), (0.0, 2.0), (-1.0, 1.0) ]) == (0.0, 1.0)

Test-driven development

Then make it happen:

def range_overlap(ranges):
    '''Return common overlap among a set of [low, high] ranges.'''
    highest_min=max([range[0] for range in ranges])
    lowest_max=min([range[1] for range in ranges])
    if (lowest_max <= highest_min): return None # Note equality
    return (highest_min, lowest_max)

Tests Define Behaviour

assert range_overlap([ (0.0, 1.0), (1.0, 2.0) ]) == None

Tests are Executable Documentation

Tests As Specifications

Testing tells you:

• If the program is doing what it's supposed to
• What the program actually is supposed to do
• Tests are runnable specifications:
• Less likely to fall out of sync with the program than documentation

Continuous Testing Infrastructure

Jenkins

Contact

@uclrcsoftdev @jamespjh @uclrits

blogs.ucl.ac.uk/research-software-development

j.hetherington@ucl.ac.uk

And sign up to be notified of the date of the next bootcamp.

``I found the command line intimidating at first, but after a while it felt like I was inside my computer.''

-- A student at the UCL software carpentry event