Bit late this week as my girlfriend was up in Nottingham for the weekend. Anyway these are the highlights from 19th Jan to 26th Jan.
- Lean Software Engineering: Software development is like golf
- Coding Horror: A visit with Alan Kay
- Code Better: Mentoring other developers
- Devlicio.us: UI patterns for WPF
- Code Better: Create your own auto registration convention with StructureMap
- Steven Harman: A sketch of our ideation pipeline
- The Runtime: Appreciating Alt.NET
- Code Better: Welcome to the anemic age
- Los Techies: Real Swiss don't need SRP, do they?
- Designer's Review Of Books: Web form design
- Los Techies: Synchronizing calls to the UI in a multi-threaded application
- Code Squeeze: Don't flaunt your best code, show us your broken crap
- Elegant Code: Stored procedure patterns
- NOOP.NL: How to handle multiple customers
- Los Techies: Ignoring testing can be explained, but never excused
- Monsters Got My .NET: You're not taking this seriously enough
Original content coming soon, I promise.
I’ve started this year wanting to improve the regularity with which I post on this blog. As part of that I’m going to record the best blog posts and articles that I’ve read over the week. This is mostly for my own reference but who knows, someone else might get some use from it to.
So here we go with what I plan to be the first of many weekly roundups.
- Code Squeeze: What C-3PO can teach you about job security
- Noop.nl: Team members, be predictable!
- Elegant Code: Ethics in software development
- Elegant Code: Please, tell me when I'm wrong. Really.
- Scott Bellware: Good design is easily-learned
- Coding Horror: If you don't change the UI, nobody notices
- XKCD: I'm an idiot
- Los Techies: Improving our industry
- Scott Hanselman: Hanselminutes 146 - Test-driven development is design
- Jon Skeet: Stack Overflow reputation and being a micro-celebrity
- Los Techies: Testing your code to test yourself
- Elegant Code: Bad apple behaviours
I’ve recently been involved in a new ASP.NET MVC project utilising LINQ to SQL as the ORM. I set up manual configuration of the mappings to try and avoid polluting the domain with all the rubbish LINQ to SQL likes to add to your domain. But a problem cropped up here that could not be overcome without tainting the domain objects: LINQ to SQL is unable to lazily load a one-to-many association without the use of EntitySet<T>. Ideally you would want to map this relationship to a List<T> or a similar data structure, but this wasn’t too bad so I did it and carried on.
Another problem with one-to-many mappings is that each child must have a property for the ID of the parent. Now this also went against my ideal as you should only have to have the parent object as a property of the child and reference the ID through that in an ideal world. Never mind, create a property for it that gets the ID from the parent object, mark it as obsolete to try and prevent people from using that property rather than the object and carry one.
Phew. This was starting to get a bit of a pain.
However, that was the end of the pain during the initial development. Everything was going pretty smoothly, people where getting to grips with LINQ queries. All was good.
But then scaling considerations reared their ugly head.
The problems we are trying to overcome revolve around returning a subset of the children through a mapping rather than the entire set. When you have a parent with a thousand children, retrieving them all just is not practical and indeed is wasteful. No-one in their right mind would try and display 1000 entries on a single web page.
Just return a subset via a LINQ query on the EntitySet<T> seemed like the easy way forward. Err, nope. EntitySet<T> doesn’t implement IQueryable so it actually retrieves every child and then uses LINQ to Objects to filter the set down to the ones you want. Hmm, what about changing the mapping in some way to filter the children there? Nope, that’s a no-go either. Seems the only option is to remove the property from the parent and retrieve the subset from a service via a proper LINQ to SQL query.
Did I mention it doesn’t do caching for you either?
Blergh, I want that domain knowledge integrated into my domain objects, not on the outskirts of it in a service. Perhaps I’m being too sensitive, but I think if you start down that path you’ll end up with next to nothing left in your domain, defeating the object of DDD. As far as I can see the only way forward is to change the ORM, which realistically leaves us with two choices. NHibernate and LINQ to Entities. NHibernate is more mature and fully featured, but would probably require rewriting all our queries to use the criteria method as there isn’t true support for LINQ as it stands. LINQ to Entities is from Microsoft (bonus to some), but it is pretty new and I’ve read about it polluting your domain quite badly.
Time to create a spike or two I think.
Posted by Garry Shutler in Unit testing
You wouldn’t want a surgeon replacing your organs without some way of knowing if he had made a mistake would you? That’s why you’re hooked up to a bunch of monitors. They’ll know the instant something goes wrong and can make educated guesses about what caused that problem.
It is exactly the same for your code.
If you don’t have a suite of unit tests giving you a reassuring green beep as you refactor a section of your system, can you really be sure that you haven’t made a fatal error?
The same analogy holds for complex changes to your system. A surgeon may well be able to remove an appendix without your vital signs being monitored; the same as you can extract the odd method. But would they even attempt heart surgery without them? Can you be sure that in altering or replacing a fundamental section of your code that you haven’t introduced problems?
A suite of unit tests give you the reassurance that any changes you made have not broken the system. This reassurance turns into confidence that enables you to make the occasional sweeping change that can dramatically improve your system; either in terms of performance or maintainability.