Remember the magic properties

You know what's a good way to feel really beaten-down at the end of the day? Spend 2 hours debugging a 1-line problem.

That's just what happened to me the other day. And, while I place some of the blame on PHP in order to make myself feel better, it was actually due to my own forgetfulness and tunnel-vision. If I had just taken a minute to step back and consider the possibilities, I probably would have had it figured out much faster.

Since last week, I have been working on the data access code for our site. Basically, we're going from using ad hoc SQL queries scattered throughout the codebase to a model-based design, where we have an abstract data model class which handles database access and then classes that inherit from that. You know, sort of an ActiveRecord pattern, a la Ruby on Rails or CakePHP.

Anyway, things were going along quite nicely. That kind of code conversion is always a little tedious, but I hadn't hit any major road-blocks and all my unit tests were passing.

Well, Tuesday afternoon, I hit a snag. I'm not sure at what point it happened, but I ran my unit tests, and suddenly I was seeing red. Lots of red. Everything was failing!

Well, I immediately when into homing mode. I started taking out tests and trying to isolate a single failure. It turns out that the first failure was in the test case for my class constructor. A little more debugging revealed the problem - none of the class members were being populated from the database. And yet, there was nothing in the error logs and the connection to the database wasn't failing.

Now, the way our (custom) abstract model class works is that it reads the table layout from the database, stores that in the object, uses it to determine the primary key of the table, and then retrieves a row based on the primary key value you pass. What was happening was that the constructor was running, getting the table data from the database (which is how I know the connection was functioning), and running the code to retrieve the row, but the values of the class members were never set.

Well, needless to say, I was baffled. Before too long, I thought I had narrowed it down to the method that actually retrieves the database row. We're using PDO (PHP Data Objects) for database access in the model class and I was using the feature of PDO whereby it will return a row from the database as or into an object. You can pass it an object and it will set each column as a field in the object. Alternatively, you can pass PDO a class name it it will return a new object of that class with the appropriate fields set from the database. You can also have it return an stdClass, which is PHP's generic object. I thought I was onto something when I discovered that fetching into the current object or fetching into a new object of the same class didn't work, but fetchin a generic stdClass did. So, of course, I eventually decided to work around the problem by re-coding the method to use the stdClass and just populate my object from that. Imagine my surprise when, even after that, and verifying that I was getting the row from the database, my unit tests still failed.

Well, to cut a long and boring story short, the problem was the magic __set() method. This is one of the "magic" functions PHP lets you define on a class. This particular one takes a key and a value and is executed every time you try to assign a value to a class property that doesn't exist. Well, we had defined that in the child class, not the model class, to do an update on the database in certain conditions and when I was re-writing it, I forgot to account for the default case - when you we don't want to update the DB, just assign the value to the key. It was one line, $this->$key = $value, which I had put inside an "if" block instead of out. So every time my code tried to do something like $this->foo = 'bar', the statement ended up being a no-op. Same thing when PDO tried to populate the object. But populating the field with the table layout worked fine, since that field was set in the class definition rather than being created at run-time.

So what's the moral of this story? Well, I guess there are several. First, tunnel-vision in debugging is a bad thing. I got hung up on the PDO object fetching and wasted lots of time trying to figure out why it wasn't working when the problem was actually someplace completely different.

The second lesson is that you need to stick to a process - in this case, I should have been doing real test-driven development. The reason this took so long is that I let far too much time elapse between the change to the __set() function and running my unit tests. If I had been doing real TDD and had run my tests after every change, I would have instantly been able to pinpoint the problem.

The last lesson is simply that software development is about managing complexity, and managing complexity is hard. Hacking away, designing as you go, relying on raw brain-power is all well and good, but it's not sustainable in the long run. When you run out of gas, or you're having a bad day, you need something to fall back on. This episode reminded me of that the hard way.

This seems to be a recurring theme for me. The longer I develop software, the more obvious it becomes that I need to imrpove my process - even whey I'm working by myself. It's kind of funny, really. The more I learn about software development, the less I feel like I know what I'm doing. There's just so much I have left to learn, it seems a little overwhelming. I know that doesn't make me inadequate - in fact, they say wisdom is knowing what you don't know. But it can still be a bit...disquieting at times.

You can reply to this entry by leaving a comment below. This entry accepts Pingbacks from other blogs. You can follow comments on this entry by subscribing to the RSS feed.

Add your comments #

A comment body is required. No HTML code allowed. URLs starting with http:// or ftp:// will be automatically converted to hyperlinks.