Sunday, April 15, 2007

How to sneak Agile into a project

Being a consultant, the opportunities I get to do project management are quite few. Still I believe many of our customers can benefit from the advantages of doing some sort of agile method. Come riding in on a white horse yelling "You are old fashioned in-effective conservatives wasting loads of effort on your specs and docs! You will go over budget if you do not join the society of Agile developers!" won't necessarily convince management that this is the way to go.

So I take some small steps to build a small proof-of-concept, merely by applying it to my own daily routine. Most developers who like to organize their own personal work in some fashion, be it writing stuff down in a notebook or filing issues in an issue tracker, will recognize these steps as mere common sense.

It can be handy to note that these routines could also be absorbed into the project on a higher level, and that is essentially when your project becomes agile.

Step 1: Ask people what they are doing and tell them what you are doing
Each morning I take a round around the office and ask "So what are you doing today?". Most people answer "Same thing I was doing yesterday, of course. Otherwise I'm going to this and that meeting". This answer points to to some of the key issues agile methods try to address (task breakdown to easier track progress).

I'll answer back "Right. Just to let you know, yesterday I completed task X. Today I'm gonna do task Y.".


Step 2: Transparent progress
I track all my tasks on the wiki. Each time I discover some new task I think needs doing, I add another item to the task-list and grade it with a complexity rating from 1 to 5 (jelly beans, story points, etc). Each week I grab a handful of tasks from the list matching the sum of jelly beans I did the week before. If one task starts taking up more than one day, there probably is some problem or obstacle I want to address to the project manager.


Step 3: Expose your problems
The number one reason for projects going wrong is task-completion being held up by dependencies, or impedements. I can not do task X before task Y is completed. To relieve waiting-time as much as possible, I maintain a list of impedements my project manager is aware of, and as long as respective impedements have not been removed, I move my progress onwards with on another task that does not have the same dependency. If you are going to blame the lack of progress on something or someone, do it as soon as possible.

Step 4: Demand short- but concrete requirements, sorted by priority
The elements in the mentioned task-list have a couple of common traits. (1) They can be completed in less than a day, and (2) they are either completed or not, i.e. they are formulated in a concrete way that leaves no half-way completion. I've either done it or I have not.

The task-list is also prioritized. I initially just pop in new tasks at the bottom of the list, but the project manager is free to move things upwards to the top of the list as he sees fit. The only thing I demand in return is that I get to complete my current task before moving on to a new one.

In the end..

People will start to pick up on my routine. Management will notice the progress, and co-workers will copy parts of the routine when they see it is working. Hopefully the concept of jelly beans or story points will quickly be shared by the team, so we can start working together on improving our weekly jelly bean consumption.

It may be that other developers are doing leaps of progress way beyond my tiny tasks, but I will claim that we are more effective when we have smaller tasks to focus at one at a time. It forces focus into your development, and you spend less time switching back and forth between various tasks, meetings and lunch.

The biggest "issue" with agile methods is that is exposes the developers' progress to the project and embraces total honesty. Traditionally, developers sometimes end up in an evil circle of reporting progress while still not having completed previous tasks, thereby pushing more and more work into "last night shifts" that can have devestating effects on both the project and the developer's personal life.

Sucumbing to total honesty can be uncomfortable for developers, as they will have to admit lack of progress and their problems, but at the same time it will get easier for the team as a whole to discover and address these problems. In the long term this will lead to a better relationship between team-members and management, and increase efficiency as all problems are dealt with as early as possible.

One final note; I call it sneaking but still your intentions should be clear to everyone. If they ask where you your ideas, say it clear: "It's a really interesting method called XP/Lean/Scrum/etc".

Got a bit side-tracked here in the end, but hopefully this is enough to get you started. Happy sneaking!

Monday, April 09, 2007

How to get off your lazy arse and contribute to an open source project

I have a bug in my code. Only it isn't I who made it. It belongs to one of the numerous open source software components in my software. How do I fix that?

I have only run into one other developer I know from the local community contributing to an open source project (not counting project devs/committers) on random occurence. Very few of open source software users contribute back to the projects. I think there is a huge amount of lurkers; people who observe the project but fail to contribute their own solutions back to the project when they encounter problems with the software (or lack of documentation).

I wrote this post while trying to get a Struts2 bug fixed. I'll keep real-life descriptions in italic font like this.

1. Identify your bug

You usually stumble on a bug when you're trying to use the component in some funny, modern, weird or alien way; integrating with another component, using rarely used/tested configuration, or simply using code from an unstable branch of the source code (bleeding edge).

The best way to isolate a bug is to recreate the conditions under which it occurs. The most handy way of writing this is by writing an executable test (a unit test if you will).

Is the bug already identified? Someone else might've had the exact same problem. Search mailing lists, issue tracker and the web in general to see if anyone else have been affected by the bug (and team up with them if so). There's no need to duplicate the identity of a bug.

I stumbled onto a bug when trying to unit-test my Struts actions' validation mechanisms. I have done so with the use of the ActionValidatorManager without problems in earlier version of Struts2 (WebWork). I searched the user-list and found another guy who has suffered the exact same problem some five days before. I added my own description of the problem to add some attention to the issue, adding some more details to the problem description, hoping that a project developer (dev) or experienced user would suggest a solution.

2. Get help

Ask around on the projects mailing list whether this is a real bug (try sticking to the user-list and not waste developer time if it's not a real issue). Many times developers or users will point out an error in your configuration, or tell you to upgrade to a later version of the component. If it is a real bug, they will probably ask you to file an issue in the issue tracker. Yes, you. Because you (the reporter) discovered the bug and you are probably the one best able to describe the problem and recognize when it has been fixed.

Three days after my mail to the thread mentioned, the problem was confirmed by Rich Thornett.

3. File an issue (document the bug)

Most issue trackers guide you into providing the crucial details you should provide to describe the bug, including environment, configuration, component version, etc. I think the best way to describe a bug is by attaching a unit-test that recreates the bug. This isn't always easy, but still there is no better way to define a bug and have a fine-grained definition of whether the bug has been fixed or not. The test is either green or red. Those of you who already do test-driven development (TDD) will recognize this step. Those of you who don't, well, get a book and figure out TDD.

Rich did us the favor of filing the issue, so I didn't have to do much at this point. I eventually (today) attached a unit-test to the issue that might get patched into the Struts2 code so the error will not reoccur in the future. I discovered that this is not really a bug if you take the effort of using the StrutsTestCase in your unit-tests, but I found a new bug being problems integrating the StrutsTestCase with the Spring objectFactory. I'll look into this later.

4. Fix the issue (submit a patch)

This is where the lazy ones quit. A project developer might be able to fix the bug in 2 minutes, so it's definitely an easy option to leave it to'em. However, if no one else suffers the effect of the bug, it will not be prioritized by the devs and it might be up to yourself to fix the bug. So get the code out! We're talking trunk-code that should be easily available from the projects SVN or CVS repository.

At this point I want to add to you anti-Mavenizers out there, this is where the advantage of Maven comes in full. I can get the code up and running in my Eclipse in 2 minutes since they use Maven. I don't have to read a friggin out-of-date developer guide before figuring out how to dev the code, spending an hour getting the deps right, then configuring them into Eclipse. Having a mavenized project makes it easier for others to submit patches to your project!

Have a look at the existing unit-tests that exist in the project. Now, for a moment (or more likely a few hours), pretend that you are a project developer and create a unit-test that isolates your bug if you didn't do so already when filing the issue). Finally, start debugging! Try to minimize the structural changes done to the source-code. This will increase the chances of the project devs absorbing your patch into the trunk-code.

In the Struts2 case, I haven't found this solution. I will continue to work on it, but hopefully it will attract enough buzz to get the devs interested and speed up the process.

Tuesday, April 03, 2007

Is Eclipse really that bad?

[Update: Stupid Blogger wysiwyg editor completely screwed up the formatting in the rss-feed.]

Patrick Lightbody asks whether my Eclipse can do that.

I'll be the first one to admit that Eclipse can be a crappy, unstable, unpredictable gob of software (especially the 3.2 release). Still his list of advantages with using IDEA strikes me as quite unfair as Eclipse can do a lot of those things he mentions.

Let's step through his points and counter what we can:

Way smarter code complete

Can go beyond the basic support provided by eclipse and can even narrow the completion down to only objects that are type-safe. That is: it won't show you a variable that, if selected, would result in a compile error.

Eclipse can do that. Hmf, hide the options that are not type correct? Eclipse suggests the variables of the correct type first, it doesn't hide the other ones, and I wouldn't want it to. What if I want to pass by that compilation error in order to change the type of a variable?

Camel case understanding

When finding classes, doing code complete, or just about anything else, IDEA understands that when you type "fUM" you really mean "fanstasticUserManager". It's a nice touch that saves a lot of typing.

Eclipse can do that. Try it for yourself.

Smart inspections

People who write code in IDEA often write code that is not as clean, and that is due to IDEA's built in smart inspections. It can do things like highlight when an "if" statement can be simplified, or when a null check will always be true.

Eclipse is halfway there with its warnings, but this might be your firmest point.

Easy to act on

Once one of these warning or error inspections pop up, IDEA makes it trivial to take action on it and fix it. Got a missing import? Alt-Enter. Have a redundant if statement? Alt-Enter. See an unused parameter? Alt-Enter. Usually IDEA will give you a few options, such as removing the offending code, changing it, or even suppressing the warning for the statement, method, class, project, or globally.

Eclipse does that. Press Ctrl+1.

It knows what you mean

IDEA is smart. It keeps track of how I name my variables and learns over time. At first, if I am creating a new variable of type BananaSundae, it'll recommend bananaSundae, sundae, and banana, in that order. But, over time, if I keep naming my BananaSundaes simply "ba", it'll start prompting that.

Eclipse will not learn from you. It will suggest bananaSundae and sundae, but not any more than that (and personally that suffices for my part).

Best HTML editor in the world

This is not an understatement. Ruby-fanatics even admit their beloved TextMate has nothing on IDEA (right Dion?). Heck, at my company, our VP of Marketing (who gets hands-on w/ web design) now uses IDEA and swears by it. Why? Because it speaks fluent CSS, HTML, and JavaScript. And all those nice inspections work here too. It'll highlight unused CSS declarations. It'll help you find usages for JS functions and CSS classes. It'll even point out the fact that "0px" is redundant and can simply be "0"... and you don't have to delete "px"... remember, just press Alt-Enter!

Remember that Eclipse by itself is a plugin-platform, and the Eclipse bundle we're talking about here is a Java-bundle, not a web-design bundle (although it does have complete CSS and HTML language support). If you wanna compare JS and HTML features, check out Aptana.

I'll accept that IDEA has a much higher rate of *just works* satisfaction. Lucky for me I've been working with Eclipse so long that I've learned to deal with its temper and can easily avoid the pitfalls, meaning poorly implemented functionality, lousy plugins, etc. But if you want to compare features and functionality head on, at least give it a fair shot and try out the features in question. Don't base it on your lousy experience with Eclipse 2.4 two years ago.

And don't compare apples with oranges. Eclipse is an open-source platform. If you want to compare it with a commercial product, atleast be gratious enough to compare it with a commercial Eclipse distribution of equal cost.

I haven't used IDEA much myself, but I've seen enough of it to believe that IDEA is a stronger tool and a better platform for the average Java developer. Still I think it's healthy that Eclipse exists as an free alternative with a much wider range of plugins.

I'm planning to run through a number of refactoring patterns with Eclipse and IDEA to see how they compare (it's actually more of a project getting to know IDEA better). More on this in a later post.