Jeff and .NET

The .NET musings of Jeff Putz



My Sites


March 2012 - Posts

Open source adventures with... wait for it... Microsoft

Last week, Microsoft announced that it was going to open source the rest of the ASP.NET MVC Web stack. The core MVC framework has been open source for a long time now, but the other pieces around it are also now out in the wild. Not only that, but it's not what I call "big bang" open source, where you release the source with each version. No, they're actually committing in real time to a public repository. They're also taking contributions where it makes sense.

If that weren't exciting enough, CodePlex, which used to be a part of the team I was on, has been re-org'd to a different part of the company where it is getting the love and attention (and apparently money) that it deserves. For a period of several months, I lobbied to get a PM gig with that product, but got nowhere. A year and a half later, I'm happy to see it finally treated right.

In any case, I found a bug in Razor, the rendering engine, before the beta came out. I informally sent the bug info to some people, but it wasn't fixed for the beta. Now, with the project being developed in the open, I was able to submit the issue, and went back and forth with the developer who wrote the code (I met him once at a meet up in Bellevue, I think), and he committed a fix. I tried it a day later, and the bug was gone.

There's a lot to learn from all of this. That open source software is surprisingly efficient and often of high quality is one part of it. For me the win is that it demonstrates how open and collaborative processes, as light as possible, lead to better software. In other words, even if this were a project being developed internally, at a bank or something, getting stakeholders involved early and giving people the ability to respond leads to awesomeness. While there is always a place for big thinking, experience has shown time and time again that trying to figure everything out up front takes too long, and rarely meets expectations.

This is a lesson that probably half of Microsoft has yet to learn, including the team I was on before I split. It's the reason that team still hasn't shipped anything to general availability. But I've seen what an open and iterative development style can do for teams, at Microsoft and other places that I've worked. When you can have a conversation with people, and take ideas and turn them into code quickly, you're winning.

So why don't people like winning? I think there are a lot of reasons, and they can generally be categorized into fear, skepticism and bad experiences.

I can't give the Web stack teams enough credit. Not only did they dream big, but they changed a culture that often seems immovable and hopelessly stuck. This is a very public example of this culture change, but it's starting to happen at every scale in Microsoft. It's really interesting to see in a company that has been written off as dead the last decade.

Rebuilding CoasterBuzz, Part II: Hot data objects

This is the second post, originally from my personal blog, in a series about rebuilding one of my Web sites, which has been around for 12 years. More:

After the rush to get moving on stuff, I temporarily lost interest. I went almost two weeks without touching the project, in part because the next thing on my backlog was doing up a bunch of administrative pages. So boring. Unfortunately, because most of the site's content is user-generated, you need some facilities for editing data.

CoasterBuzz has a database full of amusement parks and roller coasters. The entities enjoy the relationships that you would expect, though they're further defined by "instances" of a coaster, to define one that has moved between parks as one, with different names and operational dates. And of course, there are pictures and news items, too. It's not horribly complex, except when you have to account for a name change and display just the newest name.

In all previous versions, data access was straight SQL. As so much of the old code was rooted in 2003, with some changes in 2008, there wasn't much in the way of ORM frameworks going on then. Let me rephrase that, I mostly wasn't interested in ORM's. Since that time, I used a little LINQ to SQL in some projects, and a whole bunch of nHibernate while at Microsoft. Through all of that experience, I have to admit that these frameworks are often a bigger pain in the ass than not. They're great for basic crud operations, but when you start having all kinds of exotic relationships, they get difficult, and generate all kinds of weird SQL under the covers. The black box can quickly turn into a black hole. Sometimes you end up having to build all kinds of new expertise to do things "right" with a framework.

Still, despite my reservations, I used the newer version of Entity Framework, with the "code first" modeling, in a science project and I really liked it. Since it's just a right-click away with NuGet, I figured I'd give it a shot here.

My initial effort was spent defining the context class, which requires a bit of work because I deviate quite a bit from the conventions that EF uses, starting with table names. Then throw some partial querying of certain tables (where you'll find image data), and you're splitting tables across several objects (navigation properties). I won't go into the details, because these are all things that are well documented around the Internet, but there was a minor learning curve there.

The basics of reading data using EF are fantastic. For example, a roller coaster object has a park associated with it, as well as a number of instances (if it was ever relocated), and there also might be a big banner image for it. This is stupid easy to use because it takes one line of code in your repository class, and by the time you pass it to the view, you have a rich object graph that has everything you need to display stuff.

Likewise, editing simple data is also, well, simple. For this goodness, thank the ASP.NET MVC framework. The UpdateModel() method on the controllers is very elegant. Remember the old days of assigning all kinds of properties to objects in your Webforms code-behind? What a time consuming mess that used to be. Even if you're not using an ORM tool, having hydrated objects come off the wire is such a time saver.

Not everything is easy, though. When you have to persist a complex graph of objects, particularly if they were composed in the user interface with all kinds of AJAX elements and list boxes, it's not just a simple matter of submitting the form. There were a few instances where I ended up going back to "old-fashioned" SQL just in the interest of time. It's not that I couldn't do what I needed with EF, it's just that the efficiency, both my own and that of the generated SQL, wasn't good. Since EF context objects expose a database connection object, you can use that to do the old school ADO.NET stuff you've done for a decade. Using various extension methods from POP Forums' data project, it was a breeze. You just have to stick to your decision, in this case. When you start messing with SQL directly, you can't go back in the same code to messing with entities because EF doesn't know what you're changing. Not really a big deal.

There are a number of take-aways from using EF. The first is that you write a lot less code, which has always been a desired outcome of ORM's. The other lesson, and I particularly learned this the hard way working on the MSDN forums back in the day, is that trying to retrofit an ORM framework into an existing schema isn't fun at all. The CoasterBuzz database isn't bad, but there are design decisions I'd make differently if I were starting from scratch.

Now that I have some of this stuff done, I feel like I can start to move on to the more interesting things on the backlog. There's a lot to do, but at least it's fun stuff, and not more forms that will be used infrequently.

More Posts