Keep your product in ship shape

Originally ship shape meant to adhere to the standards of tidyness and alertness necessary on seagoing vessels of the old times, where things tended to get turbulent from time to time.

Nowadays, with paradigms like ship it or the famous potentially shipable product increment another meaning emerges, at least in my perception.

Don't strip the headlights to get the tail lights working

When I was much younger than today I used to repair my cars myself. Partially out of interest, partially out of distrust towards the garages and partially because of financial restrictions. And of course, because in those days it was still possible to work on ones own car.

One of the things we used to do when some part didn't work was to exchange it with something that did. (duh!)

For example one time when the headlights where not working me and a friend tried a variety of things – after exchanging the bulbs and some tweaking with the wires we finally concentrated on the fuse which looked fine. But to make sure, we exchanged it with a fuse we knew to be in working order – the one that was wired into the headlight circuit.

And sure enough the taillight lit up. So we changed everything else back to the original state – bulbs and wires and all. Or so we thought.

The one thing we did not change back was the fuse. Silly – we simply forgot.

Happy to have working taillights again (the bug was fixed) I drove home. Almost. At least as far as the next police station, where a friendly officer motioned me to stop and explained that I could not drive with a car with only one headlight.

Today I see this happening with software all the time – people fixing just the problem at hand, without realizing what havoc they bring upon the rest of the system.

So please – whenever you “fix” a piece of software: Make sure that you can still ship the whole system.

My car was definitely not shippable at that time – make your software-craftsmanship better than the car-mechanic handywork of an untrained teenaged boy and keep your product in shipshape. In a state where it can be shipped to the real customer all the time.

till next time
  Michael Mahlberg

Iterations! “Done” does not mean "never to be changed again!"

One of the hardest things to understand in agile software development seems to be the concept of “iterative“ development.

Especially in conjunction with the concepts of done-done and definition of done.

In my experience people – developers and customers alike – tend to be confused by the opposing forces of “embracing change” and getting the current iteration of the product to a “done” state in “production quality.”

Although there are several very good explanations on the web, like the one from Alistair Cockburn, it seems very hard to express the basic concepts concisely.

The default example nowadays seems to circle around the different ways to paint the Mona Lisa, so let's start with that one:

• Incremental:
  • (could be similar to waterfall with milestones)
    • partition the canvas into an arbitrary number of rectangles
    • Start painting at the top left rectangle
  • This approach has the added benefit that you could dispatch a subteam to paint the bottom right rectangle and thus ‘scale’ the project. 1
• Iterative:
  • (closer to the standard approach in agile projects)
    • Do a rough sketch of the whole picture in pencil
    • Get customer feedback on the sketch
    • Erase some lines and areas
    • Get customer feedback on the sketch
    • Do broad fills with colors
    • Get customer feedback on the intermediate
  • In this approach you might even have one iteration where the Mona Lisa was blonde

While this model does a fine job in explaining the difference between iterative and incremental, it does not really address the topics of ‘done’ or shipping.

Partially that is due to the fundamental difference between Software and pictures – it is easy to keep on working on software after shipping, because we ship something that is mechanically created from the stuff we write. (No matter whether it is compiled and linked, compiled and put in a jar, packaged or anything else – we still don't have to let go of the stuff we created to ship to customers). There is a huge difference between paintings and software, so for me this is where the analogy stops.

In painting we get value from the customer feedback. In software products we oftentimes can get much more value from real-world use of the product we’re building. Or to be exact: from using the current state of the product that we are building with real customers.

Now to be able to do this, we have to do both – we have to build an increment of our product that is fully functional. And we have to iterate over parts that we already had ‘done’ the last time around.

Let’s look at a simple example – a sign up dialog.

We might want to offer some high-tech login integration via facebook, google and twitter later on, but since we’re just a tiny startup and haven’t yet gotten around to deciphering all the fine details of the OAuth2.0 protocol, for the time being we only want to enable registration via e-mail for the first launch.

So our first iteration of the login dialog only includes e-mail. But what about the design? Since we already know that we also want Google and others integrated, we could already do a real nice UI-design for all the options, couldn't we?
Well – we could. But we shouldn't. We would have to deal with a whole bunch of decisions about things that we only have vague ideas about. And would have to maintain those decisions for quite a while. One agile way of addressing this in an iterative, incremental way, would be to build an increment that is of production quality and make sure that it is easy to re-iterate over this part easily by employing techniques from good software craftsmanship.

So the first increment would contain a finished part of the product (with all the colors, to stick with the Mona Lisa metaphor) to make it really usable, but we would be willing and take precautions to change it later on (paint it over) to iteratively add new functionality.

So long for now
  Michael Mahlberg

Footnote: Scaling and iterations

Actually this also shows that this kind of work-breakdown and scaling is manageable if the expected result to be is known beforehand. We can now easily(?) put nine illustrators on nine pieces of a gigantic billboard that is to feature the Mona Lisa as long as each illustrator knows which part they are responsible for. But that would not have worked for creating (developing) the Mona Lisa in the first place.

What's in the Potentially Shippable Product Increment (PSPI)?

There are many references to the frequent release of running software in the agile universe – of which the agile manifesto certainly gets cited the most.

One term has become very prominent - the “potentially shippable product increment” from the agile sub-universe of scrum. [In this case from LeSS - the Large Scale Scrum Framework. ’Of course’ the scrum guide itself does not use this term. The scrum guide calls it “Increments of potentially shippable functionality”]

But since the notion of the potentially shippable product increments captures so much of the practices of good agile software craftsmanship I really fell for the term and use it more often than the official term from the scrum guide.

And it is a very handy way of illustrating what is behind the ideas of Integration, Shippable, Iteratively, Product(ion quality), w

Integrated

“All [created artifacts] have to be integrated” – the stuff really has to work together (as I outlined in more detail in what exactly is continuous integration).

Shippable

Something you could give the customer to play with on their own. An installer-file, an executable, a .ear-file with deployment description, a docker image etc. Something they could really “take away” with them.

Iteratively

We know we are going to change things - let’s build them in such a way, that it is (and stays) easy to do that. Even though we deliver increments of the product we do so iteratively – we do not try to paint a picture like the Mona Lisa from the top left to bottom right. We do it with sketches and refinements and edits until we are finally satisfied with the result.

Production quality

Production quality is perhaps the hardest one to explain without painting lots of pictures. One of the best explanations I've come across is that it "also has to work when the customer can use it without the developer being around" – which implies many little things.

Those things could include:

• a user friendly way to interact with test-data if there is something like that necessary for the relevant increment
• no buttons that don’t do anything
• no spinning wheels that are not connected to any of the inner workings
• etc.

Of course this is a bad definition, because it (almost) only defines a negative-list. It only says what production quality is not. I’m still working on a positive-list, and hope to release that soon.
One thing for that positive-list that springs to mind immediately is the requirement that the code adheres to the SOLID principles – so that it only has to be changed if a future requirement has an impact on the implemented capabilities – not because something has been left over “to be fixed when we work on this in the next iteration.”

How do you define your “Increments of potentially shippable functionality?”

till next time
  Michael Mahlberg

What exactly is continuous integration?

A while ago I wrote about the fact, that a build server can't give you continuous integration, now let's look at what we might want to accomplish with continuous integration in an agile project.

Three stories and a potentially shippable product increment

Let us look at an example from a scrum-inspired project. Suppose we have three backlog items (of course not necessarily stories in the Cohn format...):

• our product can do ‘X’
• our product can be installed on Windows
• our product can be installed on Android

The ‘old’ way

In an old-fashioned project, using ‘late integration’ those requirement could easily be met by having three separate results

• a web-application where the ‘x’ can be demonstrated
• an install-package for Windows, where some sample app, based on roughly the same technology stack as the ’product to be’, is installed
• an install-package for Android, where some sample app, based on roughly the same technology stack as the ’product to be’, is installed

These different artifacts could even be put in the same source-code repository and built by some type of build server, but they would not have to interact with each other, because that could be part of a later integration phase.

The problem in this scenario is that there is no mandatory requirement for a functional integration between these tasks.

So you could end up with an installer on Windows that just displays some “this would be your application” boilerplate text, an install package for Android that contains a different launch mechanism than the windows version and a web application that can do ‘x’ but is included in neither.

Even if the automated tests related to each backlog item would be ‘green’ on the build server, the customer would still end up with three separate “features” that might not generate any value for them.

The agile way

Building this with a continuous integration mindset would be different. Following the old ideas of continuous integration in conjunction with the mindset of a potentially shippable product increment yields quite different results. Especially since such an increment contains all the existing functionality as usable and working software.

In this scenario we would focus on integrating the functionality and make sure that every story we complete interacts with all the other stories.

And all off a sudden, the customer would have real value – whether the artifacts are built via a build server or not. As long as the team creates a thing that really could be shipped every time it integrates a task – in the functional sense – and makes it easily accessible to the customer, the customer can just take that artifact(s) and interact with all of the currently completed functions.

No ci-server required, btw.

’till next time
  Michael Mahlberg