ORMs and Generated Code

At my job, we've been using a code generator for our main applications since we originally started writing them over 5 years ago. It implements Object-Relational Mapping (ORM) and generates most of the code necessary to read and write objects from the database, even includig some rudimentary business logic. Since we started, we've also extended this to include generating some things that I didn't think could even be generated, like Windows dropdowns with enumerations and a custom stored procedure for keeping a permission table up to date.

Overall, it has been a good system, and it has certainly saved us a lot of time over the years. It is very nice to be able to put a bunch of XML into a file, run the code generator, and get virtually everything we need to work with a new table or set of tables.

Still, it has some limitations, and some other issues that have made me reconsider whether using it has been a good idea 100% of the time.

First, the tool we use to generate the code was written by a consulting company that our company used for many years before the IT department became as fully staffed as it is now. Unfortunately, a couple of years ago we severed ties with that company, which means we no longer get updated versions of the tool. It was written in .NET 1.1, and the C# parser still only understands 1.1 constructs. That means if we use anything introduced in 2.0 or later, like generics, LINQ, etc., we get errors when we run the tool. We have source for most of the tool, but we are missing it for the C# parser, which is exactly what we would need to modify to fix this.

Next, I feel like in some ways, it has kept me from learning everything I could about ADO.NET. Unless we have some custom stuff we need to do in the database that the tool can't generate, we just end up calling generated methods to read and write from the database. Now, I do understand how it works, and in fact, I extended it to include the .NET 2.0 transaction model, but I still wonder if I would know more about ADO.NET if it weren't for this tool.

Finally, it has introduced somewhat of a learning curve as we have brought new developers onto the project, especially if they are not really experienced developers. Since it generates so much of the business logic and data access layers, developers have to be trained to not just jump in and start writing or changing code that touches those areas. They need to understand that some of their changes may be wiped out by the tool if they don't do them correctly. So far, most have been positive about this, and have caught on quickly, but it is still something different than most people are used to.

So, looking back, would I have done anything differently if I could have? It's hard to say. Probably the biggest change I would have liked to have made would be to somehow keep the generated code separate from any custom code we wrote. If we had had .NET 2.0 back then, we could have used partial classes (and in fact, if I had the source to the C# parser, that would be the first thing I would add). As it is, the code is all mixed together, and it can sometimes be hard to tell what is getting generated and what is hand-coded. I definitely would not have chosen to write everything by hand. I definitely believe generating code like this is beneficial.