Half a Trade, half an Art

Configuring Eclipse Subversion plugins

Entia non sunt multiplicanda praeter necessitatem.

William of Occam

Don't make life difficult for yourself.

Me (paraphrasing badly)

All I wanted to do was check out some Java!

Some backstory - we have a subversion repository holding our source code, which we connect to over ssh for security.

My first problem was that I couldn't run ant scripts that checked out code from subversion. I use cygwin and openssh for scp and ssh normally. The ant task for subversion uses the JavaHL library if it can find it. It can on my machine, because one of the many subversion client installations has installed it.

The build would hang at the svn task. This was happening because my ssh keys were locked with a passphrase. The svn task can't handle this, and I didn't have time to hack in support. So I removed the passphrase. This isn't good, but practically I don't need the protection as much as I need to be able to run the script.

So now the ant script could check out components and build a deployable server. But the next problem was that it couldn't copy it over to the target host. The task didn't have access to a list of known host signatures. There are two options here - trust any host that sits at the correct IP address, or provide a list of known hosts. I had such a list - openssh builds one up for me. But I couldn't hardcode my own list into the build script - it wouldn't work for the other users. I put a copy of the list into the project instead.

This worked, but my next problem was using Eclipse as well. I like to run ant scripts from the command line, but check out modules and do development using Eclipse.

I installed the subversive plugin for Ganymede because it is meant to be nicer to use than the subclipse plugin that comes pre-installed. I installed the SVNKit and JavaHL connectors from the installation site, but only the SVNKit connector would work.

The weird problem with all of this is that although I could check out the release module in Eclipse, it would come out in SVNKit format. When I ran the script from the command line it would check out components into subdirectories using JavaHL. JavaHL uses a format SVNKit cannot read. If I refreshed the project in Eclipse, it detected this and forced me to switch to using the JavaHL connector - which won't work. So I had to hide my command line activities from Eclipse, which was going to drive me mad.

It turns out that subversive uses JavaHL provided by subclipse - so you configure it the same way for everything. What was causing the trouble for me is that under cygwin this is a directory in /home/adrian/.subversion but under ant and any windows applications this is in my application data directory in a subversion directory. The command line worked because the default tunneling app for ssh was available under cygwin - but there wasn't an ssh available on the path for windows. Once I fixed the config in the application data folder everything was working.

Except for a small problem. My openssh executable popped up a dos window for every call. There are quite a few calls happening in Eclipse. I had to switch to the TortoisePlink.exe in the TortoiseSvn installation instead. This doesn't pop up a dos window - it just pops up a window asking you to log in. To stop this you have to give it the location of a private key file. You can't use your openssh key file, though - that would be too easy. Private key files don't have a standard format, so you have to convert the openssh keys into Putty format first. This involves installing the Putty key generator. Then you supply the location of the Putty keyfile as a -i option to the plink command.

Finally, this fixed Eclipse so that I could use the JavaHL connector everywhere and everyone was happy.

Easy.

Unbricking an Airport Extreme

How I unbricked my Airport Extreme, for all the other people having difficulty with firmware upgrade 7.3.2.

A couple of days ago my Macbook told me that there was a firmware upgrade for my Airport Extreme (802.11n) - and like a fool I installed it, as prompted.

I was lulled into a false sense of security by the fact that the Airport management software seemed completely happy to try a firmware update over the air. Netgear firmware updates always gave big warnings. Even over the wire, they made clear that firmware updates were like doing your own brain surgery. I half-consciously assumed that Apple had implemented a transactional firmware update mechanism.

As the 'waiting for your Airport to reboot' window lingered, I became more and more aware that I had done a Bad Thing. The Airport didn't appear in the list of managed devices. It didn't appear in my WLAN list. The light cycled between green (for a minute or so), solid amber (for some time) and flashing amber (for a minute or two). I don't remember the order that the lights changed etc.

I tried cycling the power. Waiting for capacitors to drain. Same behaviour.

I tried factory reset. Same behaviour.

I tried factory reset while powering up. Same behaviour.

I reinstated the old WLAN (on the ADSL router) and checked the web. Some people had commented that powering up some time later fixed the problem.

I powered up the next day, and the behaviour was different. What changed? Maybe I just cycled power enough times to reach a change in behaviour. Maybe the Airport has a battery and needs longer to drain than the 10 minutes for capacitors. Fully drained, the box may have had a different startup cycle. In any event, it still didn't work, but the light was on solid amber.

My ADSL router has lights on the front showing which ethernet ports at the back are connected to working ethernet connections. The only plug being used was the one for the Airport, and the light was dark. I checked the ethernet port on the back of the Airport, and the two little lights were dark. The Airport WLAN (Internet) NIC had not come up.

I didn't think about it - I just moved the ethernet cable from the WLAN port to one of the other ports on the back of the Airport. This is usually a Bad Idea because you ought to only use the WLAN port for the Router. The NIC behind the port will be heavily firewalled, to protect the Airport (and the machines using it for connectivity) from the mad, bad Internet. The non-WLAN ports are much less heavily protected, as they should only connect your own trusted hardware to the Airport. I had just plugged the entire world directly into my Airport.

Not to worry, though - my Router has a separate firewall in front of the Internet, so nothing got through. I should have thought more about it, though.

With the router plugged in via one of the DMZ ports, suddenly a light came on showing an active ethernet connection - and the DMZ port's little lights came on. And the Airport light went beautiful, solid green. At this point I realised that I really ought to have plugged it only into the WLAN port - and moved the ethernet cable back. The WLAN port's lights came on - so in between shifting the cable, the Airport had brought the WLAN NIC up. I checked that everything was actually correct by rebooting - and indeed the Airport was fine.

Interestingly, even following several factory resets, all my configuration information - security settings, passwords, NAS setup - were all intact.

So it seems like the firmware update hit some sort of deadlock. It needed network access before it could bring up the WLAN NIC.

Anyway, if anyone out there has a similar problem, it might be worth doing the same things. Just make sure you are safe plugging your router into one of the DMZ ports first.

Generic Arrays...

I just found out that you can't create a generic array in 1.5. What? This isn't allowed:

Satck<Date>[] dates = new Stack<Date>[69105];

Instead, you have to do this:

Stack<Date>[] dates = (Stack<Date>[])new Object[69105];

Generics keep surprising me. Overall it is good to be able to do template metaprogramming, but they totally broke the rule of least surprise. I want to say that they shouldn't have maintained backwards compatibility - but then I guess we wouldn't have got generics at all.

If the unchecked casting warnings annoy you, you will want this annotation:

@SupressWarnings("unchecked")

Who wrote that?

Ever impressed by code, only to find out that you wrote it - and feel that it is now above your level? Here is a link to some code that gave me that feeling.

I posted because JBoss completely changed the way remote JMX worked when upgrading from version 3 to 4.  There weren't even any hints in the documentation. My questions on the boards went unacknowledged, never mind answered.

Eventually I grokked enough RMI to sort it out myself, and posted the results. The post is overlong because it is a complex thing to set up - and in fact I wrote it and don't follow it now. Some people have found it useful though, because I received e-mails thanking me for posting it!

I seem to remember (unreliably) that the post was removed pretty quickly, and the right way to do it was posted instead. I was slightly annoyed because the comments effectively said we should have looked at the unit tests for guidance. Now really, there are two things wrong with this. Firstly, unit tests are not good user documentation - they deal with too much detail for a start. Secondly, a lot of us use a binary distribution (no tests included) and expect the documentation to explain things. Even so, I was annoyed at myself too, for not checking the tests after all.

JBoss have a bit of a reputation for responding to questions in an unpleasant way, so I was quite surprised to see my post re-instated a little later. Someone obviously had a change of heart, which was nice. Both approaches were there on the wiki, and there they have stayed.

I still don't follow the code, though.

Hello Technorati!

Technorati Profile

Mission Statement

This is somewhere to record my ramblings about computer science and software development.