Hello App Engine

Yesterday morning I saw the news that Google App Engine was now open for Java as well as Python applications – in beta, that is. I signed up and received notification of access almost immediately. I read the notes on getting started with Eclipse. Fortunately I already have Eclipse installed. I just needed to run Eclipse, and enter the URL of the Google plug-in into the software update configuration dialog. The plug-in downloaded and installed in a few moments. Then it is a matter of File – New – Other and select Google Web Application Project. Enter a project and package name, click Finish.

The wizard creates a skeleton Java Servlet application. I made a few trivial modifications to both the servlet code and the home page. Clicked Run, and the app runs on a local server. It worked. Next, I needed to deploy it. I signed into my App Engine console, and created a new application. I had to find a name that was not yet taken, and selected javaisgo. This generates an application ID. I copied this into the appengine-web.xml file which the Eclipse wizard had generated. Then I hit Deploy App Engine Project in the toolbar. I was prompted for my Google account name and password, the application uploaded, and it was done; you can see the results at http://javaisgo.appspot.com/.

Although Python is dynamic and fashionable, Java is probably the world’s most popular language for business development. I expect the ability to create and deploy applications so easily and for free will be attractive to many, leaving aside anxiety about Google’s plans to take over the Internet.

Hello Windows Azure

All this reminded me that although I had signed up for the Windows Azure CTP (Community Tech Preview) a while back, I had not got round to deploying a web application. I did deploy a Mesh-Enabled Web Application, part of the Live Framework, a process which I found frustrating and ultimately disappointing. That is a different kind of thing though; whereas an Azure ASP.NET application takes a similar approach to that of Google App Engine – write your web application, deploy to the cloud provider’s servers.

I already had an Azure account and developer token; how long would it take to deploy a Hello World project to Azure? Well, first you have to install the Windows Azure Tools for Visual Studio. I tried running this, but the dependencies were not in place. Although I already had Visual Studio 2008 SP1, SQL Server, and .NET Framework 3.5, I needed to add IIS 7.0 with ASP.NET to my development machine, and to configure the .NET Framework to support WCF HTTP Activation, which is off by default. See here for details. I did all this and the tools installed. The SDK also gets installed. When you build and run the samples, it starts up two new services on your machine, Development Fabric and Development Storage.

Next, I started Visual Studio, which apparently has to be run with full local admin rights for Azure to work. New project, Visual C#, Cloud services, Web Cloud Service (essentially an ASP.NET application). This looked familiar, and I quickly added a button and an event handler to make sure it worked. When you debug, it runs on the local development fabric.

Time to deploy. This is where I ran into some difficulties. I logged into the development portal and created a new Hosted Services project, called Azure is go. Next, I went back to Visual Studio and used the Publish wizard. Note: you must not use Publish from the Build menu, as this does not work. You need to right-click the solution in the Solution Explorer and choose Publish from there. This compiles two files, a .cspkg which contains your application, and ServiceConfiguration.cscfg which is configures it.

The wizard is disappointing: it merely opens an Explorer window showing your deployment files, and opens the Developer Portal in a web browser. You deploy your project to a Staging area; then when you are happy with it, hit Promote to copy it to the production URL. In order to deploy, you have to select the two deployment files manually in a web upload dialog. The reason the Wizard opens the Explorer window is to show you where they are so you can copy the path. All rather clunky, though not difficult.

After I did this, the Developer Portal displayed a spinning bagel and the words Package is deploying; then eventually it said Allocated. I hit Run, and it said Initializing. Nothing seems to happen quickly with Azure. It said Initializing for a long time; I got bored and hit Refresh. This may have been a mistake. My Staging icon went red and I got a message: InternalServerError – Information is not available.

I decided to delete the deployment and retry. I clicked Delete, whereupon Azure told me it couldn’t delete the deployment because “tenant status is currently Running”. That almost seemed hopeful; yet the test URL still did not work. I clicked Stop – pause while it stops – then Delete – pause while it deletes – then re-compiled and re-deployed.

My second effort worked, after the usual pauses. I then selected Promote and the application arrived at its final URL: http://azureisgo.cloudapp.net. As with App Engine, it is good to know that your web site is running on a scalable data center rather than on a single machine or virtual machine, and currently without any cost.

These hello world experiences may not seem important later, when you are buried in the intricacies of a real application, but they do have an impact on your desire to explore and experiment with a new platform. Judging from my own experience, getting started with Google App Engine is easier than it is with Azure, even for a Windows developer already set up with Visual Studio. The long pauses as Azure thinks about deploying your project also make a bad impression, in contrast to App Engine’s near-instant response. Maybe it will all come right with Visual Studio 2010 and the final release of Azure. In the meantime, this does nothing to shake my feeling that Microsoft’s Azure launch needs attention if it is to win developer mindshare.

I’m mostly using Vista 64-bit these days and enjoying it. The system is stable, fast and responsive. At least, it was until I started using a dual display; this morning I got a blue-screen, which Windows assured was because of the display driver. I also noticed that the Windows Problem Reports and Solutions applet said that my Intel network driver was causing problems and should be updated, so after updating the NVidia graphics driver, I also downloaded the latest from Intel and tried to run it.

It wouldn’t install. It got almost to the end, then declared:

Error 1713. Intel Network Connections cannot install one of its required products. Contact your technical support group.

The event log also refers to System Error: 1605. This means: This action is only valid for products that are currently installed.

The solution is to uninstall the existing Intel networking utility. Not easy, since bizarrely the uninstall fails with a report that the software is not designed for this version of Windows. So why did you let me install it then? And why not let me remove it?

The next step is to use msizap, also known as the Windows Installer Cleanup Utility, to delete the installation from the Installer database. Note that it doesn’t install any actual files. I downloaded the cleanup utility here, but it did not work; when I tried to run it, the GUI did not appear. That download seemed quite old, so I installed the more recent Windows Installer 4.5 SDK which includes an updated msizap. That version is command-line only, but I copied the new msizap to the Windows Installer Clean Up folder in Program Files (x86); then the GUI worked too. Removed the Intel networking utility, re-installed the updated driver, and all is well.

Thank also to this thread for some useful pointers.

I’ve been messing around with web form design recently; started with a table layout, decided it was horrible and unmanageable so redid it with CSS. I came across this example which had more or less the layout I wanted – it’s the form about half way down that looks like this:

Looks simple enough; and is based on an idea by CSS guru Eric Meyer; so I copied the code and tried it. Unfortunately my version looked like this:

I squinted at the code again and noticed that a style defining a little hack called div.spacer was missing from the sample code though it is mentioned earlier in the article. Added it, and now my form looked like this:

Better, but frankly not that close to what I wanted. What was wrong? Should the fields float:left instead of float:right? It made little difference. Then I noticed that the example on the page was real code, not just an image. I downloaded the actual CSS it was using. Of course it was an obvious error. The sum of the widths assigned to the label style and the field style was greater than the width of the containing div. Increase that, and all is well:

Great stuff; but it reminded me how tricksy CSS is. With both cascades and inheritance, and exceptions such as the fact that some properties do not inherit, figuring out exactly what style attributes apply to an individual element is a challenge. The positioning rules are complex and often do not work as I first expect. Styles can be defined in numerous places; and while external CSS files are easier to manage than those defined within HTML, they soon get long and hard to navigate.

By way of mitigation, CSS is powerful; yet as the above example shows, it can still need workarounds (div.spacer in this case) to achieve the result you want.

I suppose I’m waiting for that aha moment when it all makes perfect sense; but it seems a long time coming. Worried that it was just me, I found this reassuring post from Sho Kuwamoto:

I used to think of myself as knowing a lot about CSS. For starters, I’d been responsible for the CSS implementation in Dreamweaver. I was also a member of the W3C CSS working group. I wasn’t a major contributor (I didn’t author any of the chapters of the spec, for example), but I thought I knew the spec pretty well.

It’s been a while since I’ve touched CSS, and in coming up with the design for this blog, I was reminded again how difficult it is to use CSS to get the layout you want. It was incredibly difficult. I couldn’t get it to work and I ended up having to google around to figure out how other people had done their page layouts.

I’ve also noticed that the aforementioned Eric Meyer is increasingly critical of the language. In his post Wanted: Layout System he writes:

Maybe CSS isn’t the place for this. Maybe there needs to be a new layout language that can be defined and implemented without regard to the constraints of the existing CSS syntax rules, without worrying about backwards compatibility. Maybe that way we can not only get strong layout but also arbitrary shapes, thus leaving behind the rectangular prison that’s defined the web for almost two decades.

It’s too late of course. Now that Microsoft’s Internet Explorer 8 is out there is decent support for CSS across all the major browsers. What’s the chance of getting agreement on a new layout system now? The only realistic alternative is to work increasingly in Adobe Flash or Microsoft Silverlight, which is proprietary badness but can be attractive.