Distributed and agile development in PyPy

Note

This page describes the mode of development that preceeds the current models of Open Source development. While people are welcome to join our (now yearly) sprints, we encourage engagement via the GitHub repo at https://github.com/pypy/pypy/. Issues can be filed and discussed in the issue tracker and we welcome pull requests.

PyPy isn’t just about producing code - it’s also about how we produce code. The challenges of coordinating work within a community and making sure it is fused together with the parts of the project that is EU funded are tricky indeed. Our aim is of course to make sure that the communities way of working is disturbed as little as possible and that contributing to PyPy still feels fun and interesting (;-) but also to try to show to the EU as well as other funded projects that open source ideas, tools and methods are really good ways of running development projects. So the way PyPy as a project is being run - distributed and agile - is something we think might be of use to other open source development projects and commercial projects.

Main methods for achieving this is:

  • Sprint driven development
  • Sync meetings

Main tools for achieving this is:

  • py.test - automated testing
  • Git - version control
  • Transparent communication and documentation (mailinglists, IRC, tutorials etc etc)

Sprint driven development:

What is a sprint and why are we sprinting?

Originally the sprint methodology used in the Python community grew from practices within Zope3 development. The definition of a sprint is “two-day or three-day focused development session, in which developers pair off together in a room and focus on building a particular subsystem”.

Other typical sprint factors:

  • no more than 10 people (although other projects as well as PyPy haven been noted to have more than that. This is the recommendation and it is probably based on the idea of having a critical mass of people who can interact/communicate and work without adding the need for more than just the absolute necessary coordination time. The sprints during 2005 and 2006 have been having ca 13-14 people per sprint, the highest number of participants during a PyPy sprint has been 24 developers)
  • a coach (the coach is the “manager” of the sprint, he/she sets the goals, prepares, leads and coordinate the work and track progress and makes this visible for the team. Important to note here - PyPy have never had coaches in our sprints. Instead we hold short status meetings in the whole group, decisions are made in the same way. So far this have worked well and we still have been able to achieve tremendous results under stressed conditions, releases and such like. What we do have is a local organizer, often a developer living in the area and one more developer who prepares and organizes sprint. They do not “manage” the sprint when its started - their role is more of the logistic nature. This doesn’t mean that we wont have use for the coach technique or something similar in the future).
  • only coding (this is a tough one. There have been projects who have used the sprinting method to just visionalize och gather input. PyPy have had a similar brainstorming start up sprint. So far though this is the official line although again, if you visit a PyPy sprint we are doing quite a lot of other small activities in subgroups as well - planning sprints, documentation, coordinating our EU deliverables and evaluation etc. But don’t worry - our main focus is programming ;-)
  • using XP techniques (mainly pairprogramming and unit testing - PyPy is leaning heavily on these aspects). Pairing up core developers with people with different levels of knowledge of the codebase have had the results that people can quite quickly get started and join in the development. Many of our participants (new to the project and the codebase) have expressed how pairprogramming in combination with working on the automated tests have been a great way of getting started. This is of course also a dilemma because our core developers might have to pair up to solve some extra hairy problems which affects the structure and effect of the other pairs.

It is a method that fits distributed teams well because it gets the team focused around clear (and challenging) goals while working collaborative (pairprogramming, status meeting, discussions etc) as well as accelerated (short increments and tasks, “doing” and testing instead of long start ups of planning and requirement gathering). This means that most of the time a sprint is a great way of getting results, but also to get new people acquainted with the codebase. It is also a great method for dissemination and learning within the team because of the pairprogramming.

If sprinting is combined with actually moving around and having the sprint close to the different active developer groups in the community as well as during conferences like PyCon and EuroPython, the team will have an easier task of recruiting new talents to the team. It also vitalizes the community and increases the contact between the different Python implementation projects.

As always with methodologies you have to adapt them to fit your project (and not the other way around which is much too common). The PyPy team have been sprinting since early 2003 and have done 22 sprints so far, 19 in Europe, 2 in the USA and 1 in Asia. Certain practices have proven to be more successful within this team and those are the one we are summarizing here.

How is it done?

There are several aspects of a sprint. In the PyPy team we focus on: 1. Content (goal) 2. Venue 3. Information 4. Process

  1. Content (goal) is discussed on mailinglists (pypy-dev) and on IRC ca one month before the event. Beforehand we have some rough plans called “between sprints” and the sprintplan is based on the status of those issues but also with a focus on upcoming releases and deliverables. Usually its the core developers who does this but the transparency and participation have increased since we started with our weekly “pypy-sync meetings” on IRC. The sync meetings in combination with a rough in between planning makes it easier for other developer to follow the progress and thus participating in setting goals for the upcoming sprints.

    The goal needs to be challenging or it won’t rally the full effort of the team, but it must not be unrealistic as that tends to be very frustrating and dissatisfying. It is also very important to take into account the participants when you set the goal for the sprint. If the sprint takes place connected to a conference (or similar open events) the goals for the actual coding progress should be set lower (or handled in another way) and focus should shift to dissemination and getting new/interested people to a certain understanding of the PyPy codebase. Setting the right goal and making sure this is a shared one is important because it helps the participants coming in with somewhat similar expectations ;-)

  2. Venue - in the PyPy project we have a rough view on where we are sprinting a few months ahead. No detailed plans have been made that far in advance. Knowing the dates and the venue makes flight bookings easier ;-) The venue is much more important than one would think. We need to have a somewhat comfortable environment to work in (where up to 15 people can sit and work), this means tables and chairs, light and electricity outlets. Is it a venue needing access cards so that only one person is allowed to open? How long can you stay - 24 hours per day or does the landlord want the team evacuated by 23:00? These are important questions that can gravely affect the “feel and atmosphere” of the sprint as well as the desired results!

    Also, somewhat close to low cost places to eat and accommodate participants. Facilities for making tea/coffee as well as some kind of refrigerator for storing food. A permanent Internet connection is a must - has the venue were the sprint is planned to be weird rules for access to their network etc etc?

    Whiteboards are useful tools and good to have. Beamers (PyPy jargon for a projector) are very useful for the status meetings and should be available, at least 1. The project also owns one beamer - specifically for sprint purposes.

    The person making sure that the requirements for a good sprint venue is being met should therefore have very good local connections or, preferably live there.

  3. Information - discussions about content and goals (pre announcements) are usually carried out on pypy-dev (mailinglist/IRC). All other info is distributed via email on pypy-sprint mailinglist and as web pages on codespeak. When dates, venue and content is fully decided a sprint announcement is being made and sent out to pypy-dev and pypy-sprint as well as more general purpose mailing lists like comp.lang.python and updated on codespeak - this happens 2-4 weeks before the sprint. It’s important that the sprint announcements points to information about local transportation (to the country and to the city and to the venue), currency issues, food and restaurants etc. There are also webpages in which people announce when they will arrive and where they are accommodated.

    The planning text for the sprint is updated up till the sprint and is then used during the status meetings and between to track work. After the sprint (or even better: in between so that the memory is fresh) a sprint report is written by one of the developers and updated to codespeak, this is a kind of summary of the entire sprint and it tells of the work done and the people involved.

    One very important strategy when planning the venue is cost efficiency. Keeping accommodation and food/travel costs as low as possible makes sure that more people can afford to visit or join the sprint fully. The partially EU funded parts of the project do have a so called sprint budget which we use to try to help developers to participate in our sprints (travel expenses and accommodation) and because most of the funding is so called matched funding we pay for most of our expenses in our own organizations and companies anyway.

  4. Process - a typical PyPy sprint is 7 days with a break day in the middle. Usually sprinters show up the day before the sprint starts. The first day has a start up meeting, with tutorials if there are participants new to the project or if some new tool or feature have been implemented. A short presentation of the participants and their background and expectations is also good to do. Unfortunately there is always time spent the first day, mostly in the morning when people arrive to get the internet and server infrastructure up and running. That is why we are, through documentation, trying to get participants to set up the tools and configurations needed before they arrive to the sprint.

    Approximate hours being held are 10-17, but people tend to stay longer to code during the evenings. A short status meeting starts up the day and work is “paired” out according to need and wishes. The PyPy sprints are developer and group driven, because we have no “coach” our status meetings are very much group discussion while notes are taken and our planning texts are updated. Also - the sprint is done (planned and executed) within the developer group together with someone acquainted with the local region (often a developer living there). So within the team there is no one formally responsible for the sprints.

    Suggestions for off hours activities and social events for the break day is a good way of emphasizing how important it is to take breaks - some pointers in that direction from the local organizer is good.

    At the end of the sprint we do a technical summary (did we achieve the goals/content), what should be a rough focus for the work until the next sprint and the sprint wheel starts rolling again ;-) An important aspect is also to evaluate the sprint with the participants. Mostly this is done via emailed questions after the sprint, it could also be done as a short group evaluation as well. The reason for evaluating is of course to get feedback and to make sure that we are not missing opportunities to make our sprints even more efficient and enjoyable.

    The main challenge of our sprint process is the fact that people show up at different dates and leave at different dates. That affects the shared introduction (goals/content, tutorials, presentations etc) and also the closure - the technical summary etc. Here we are still struggling to find some middle ground - thus increases the importance of feedback.

Can I join in?

Of course. Just follow the work on pypy-dev and if you specifically are interested in information about our sprints - subscribe to pypy-sprint@codespeak.net and read the news on codespeak for announcements etc.

If you think we should sprint in your town - send us an email - we are very interested in using sprints as away of making contact with active developers (Python/compiler design etc)!