dasman’s World

Its the new development methodology thats taking the developer-world by storm….

Well, not really.

Just a cute phrase that occurred to me today when chatting to a colleague. We’d just spent 15 minutes discussing the desired behaviour of a system I was working on. Usual stuff – at first blush it seems straightforward, but plenty of edge cases, historical behaviour requirements, and other nastiness.

So we got it figured out, and its 11:45. Lunch booked at 12:00. No way am I even getting 10% of the code done in that time. But its just enough time to do the Javadoc.

So, 15 minutes later, my code has the 2 method signatures I need and around 40 lines of Javadoc describing the code’s behaviour, exceptions, and assertions. And after lunch, I was able to get straight into coding, all details of my discussion safely preserved.

And as an added bonus, when it comes time to rework it, I (or the next guy) will have some great doco to get started.

Not rocket science. Certainly not worthy of its own buzzword-esque title (although it does roll off the tongue ). And probably something everybody should be doing anyway as part of their actual methodology. But its a neat little anecdote worth telling!

Well, they say you learn something new everyday. And given the size of the Java API, thats not too hard to believe

After doing serious Java programming for at a least a few years, and having dabbled for a few years more (all the way back to the 1.0 days!), I’m pretty happy with my breadth of Java knowledge. Until a co-worker today mentioned using Collections.shuffle() to randomize a list.

“Pffffft”, says I, “Surely you jest”, says I. “No such thing.”

Well, there is. java.util.Collections (not Collection, note – theres an “s” on the end!). A nice, chunky class, full of juicy, static methods just waiting to save you from having to re-invent the wheel when it comes to operating on Collections. There’s the obvious stuff – shuffle() and sort() to randomize and sort, for instance. And small, but useful stuff – min() and max(), swap(). And some real useful stuff, like synchronizedList() (which returns a thread safe version of your list).

Many’s the time I’ve done my own “max()” routine. Each time, I trawled through the Collection interface looking for one and was mildly surprised it didn’t exist. And I never thought to look any further afield. I’m guessing that there may be one or two Java guys out there who are in the same boat (How many library classes are there now in Java?). Seeing as my co-worker doesn’t have his own blog…yet (come on H! Get your act together! ), I’m doing this posting for the other people like me!

Of course, I’ve always got the excuse for my ignorance that its a relatively recent addition. I mean, its only been around since Java 1.2….. *grin*

I came across this entry while scanning Java Blogs yesterday about how to handle exceptions within a constructor. The essence of the article is how do you handle the situation where you couldn’t instantiate your object?

So, if the user passes invalid parameters, do you throw an exception? And, slightly more subtly, if your constructor encounters an exception during object setup, how does it handle this?

Interesting article and well worth the read.

But it struck a bit of a chord with me – I recently bought a Nokia N70 phone, and have been cruising the net looking at how you can code for these things. The N70 is a Symbian phone, and while it has J2ME support, the hardcore stuff seems to be done in C++. Now, C++ is an … interesting language, but is not really known for its in-built protection against memory leaks, and when you’re running a program on a device with limited memory, where your app may well run for the order of days or months this suddenly becomes a big deal.

OK, so you just code, really, really carefully.

Well, turns out that constructors are a special case in terms of memory leaks. Symbian doesn’t use standard C++ exceptions, but it does have an analagous concept called “leaving” grafted on. And not being a C++ guru, I couldn’t tell you what difference there is between normal C++ exceptions and the Symbian variety. In any case, the issue when writing Symbian programs appears to be this:

1. Constructor called
2. Constructor initialises an instance variable that points to some new object
3. On the next line, the constructor encounters an error and leaves

So, we now have allocated memory for some new object (step 2 above), but the object that owns it no longer exists – it crashed during construction. And because the object never fully formed, the destructor wasn’t run on it and so couldn’t free up the memory. Ladies and gentlemen, we have memory leak….

Turns out that Symbian programs have an accepted idiom on how to handle this stuff. Objects that can suffer this way have a static construction method (like the whole getInstance() idiom you see in Java). This static constructor creates the object using a constructor that only performs trivial operations that won’t crash, then once the object exists, calls an instance method on the new object that performs 2nd-phase construction. If things turn to jelly, there is at least a concrete object in existence, so the destructor will be called as normal.

Pretty neat. And pretty eye-opening as to the kind of issues you can be exposed to without garbage collection.

(Which is not to say that Java doesn’t have its own issues with memory leaks, but thats a whole other blog post… )

A week or so ago, I downloaded NetBeans 5 RC2 specifically to play with the mobility pack, but figured I might as well explore its capabilities as a J2SE IDE.

For those of you who like their blog posts short and snappy – “Neat, some cool stuff in there, but I’ll be sticking with Eclipse for now”. For those who can handle a couple more paragraphs, read on….

To start with, can I join the crowd and say “Matisse rocks!”? Matisse is the visual editor for Swing apps, and it is fantastic – in 90% of cases it does what you mean: you put an object right on the left border, and it figures that that object should be anchored to the left. You place one button next to another and it keeps them together and snaps them to be level. I played with Eclipse’s Visual Editor in a recent small project, and it has a ways to go to catch up to this.

But (and there’s always a “but”, right?), the killer for me was that “90% of cases” phrase – its the other 10% that comes back to bite you. For instance, I wanted to do a a little “cross” of buttons – and up, down, left and right, laid out in a diamond. After having been spoilt with the editor doing what I meant for the rest of the UI, I found it a bit jarring when I was doing “free form” component layout – buttons were slightly off, by a pixel or two and it kind of “looked” wrong.

There’s probably a hidden “layout in a cross” option I missed . And I could have added a panel, set it to GridBagLayout and done it that way (Matisse has good support for GridBag). And how often do you want to have buttons in a bizarre shape like that anyway? But it kind of broke the spell for me in a way. I’m a big believer of being “in the zone”, development-wise, and switching between IDE modes (code and visual) is a real context switch for your brain – to compensate, I think I need almost perfection from the Visual Editor. Matisse is soooooo close, but isn’t quite there for the fiddly cases. (How would I have liked Matisse to handle my peculiar layout requirement? I have no idea!).

Matisse is the big ticket item – the rest of the IDE is good, but not compelling from where I sit. There are some winners, though. I love the idea that the IDE is ant based – take your development directory, zip it up and give it to someone else, and they can compile, test, build, etc. Ant seems to be the one big constant across teams – it makes so much sense to have it as the backbone of your development processes. And the fact that a test folder is created when you start a new project is a little nicety, but a good one.

On the downside, on my machine (2 Ghz with a ridiculous 512Mb of RAM), NetBeans is noticeably more sluggish than Eclipse – maybe this is an SWT thing? (I hope not, I’m one of those Swing fans ). And the plugin environment for Eclipse seems so much richer at the moment – I even have a VI plugin for my Eclipse for when I’ve been doing a lot of Unix work and keep on putting the letter “i” throughout my source code (you Unix guys will know what I mean)!

So, I’m staying with my Eclipse for now. It was certainly worthwhile to check out though, and I’m still planning to do my next little J2ME project with NetBeans – I’m keen to give that Mobility Pack a proper workout.

There’s a worthwhile article on code reviews at developer.com at the moment called Effective Code Reviews Without the Pain.

Its a quick read, and the title says it all. The article’s focus isn’t so much on what your code review should cover, but how your code review should be carried out – how to make what can feel like a very adversarial, negative process into a postive, learning process. There’s tips for both the code reviewer and reviewee, with points like “Avoid ‘Why’ questions” and “Remember that the code isn’t you”.

Good, common-sense stuff. Its rare to find a development area that doesn’t use code reviews nowadays (even if only intermittently! ), so it should be worth a read for most developers.