Illusory Follies Andrew Flanagan's Blog


On Mountains and Software

Way up in the hinterlands, miles from any other humans, two men faced each other on a rocky ridge covered in lava flows and avalanche lilies.

“It was clearly a suboptimal algorithm!” the taller man yelled, with his arms outraised.

“Everyone knows that premature optimization is the root of all evil” the short, stouter man parried. “If we had known all the details of the operational environment at the time that we had implemented the original code, we might have made different decisions, but ultimately, it doesn’t matter. The code worked and the solution was a good one -- at least for the first iteration.”

“The first iteration?! What is with you and iterations of software releases? Why not spend the time to do it right on the first try. Now your code is out there and it’s ridiculously slow -- the whole company looks bad when you deliver crap like this.”

The short man leaned against a large basalt pillar that rose almost unnaturally out of the mountain, like an ancient stella, erected to point out the glory of this lofty pinnacle. “First, make sure it works, then make it fast.” he said slowly. “Let me ask you this, do you think that the company would have looked better or worse if we had spent an extra 2 or 3 weeks on the initial release to make it faster but the whole thing had a fundamental logic error?”

The tall man writhed, “Why do you assume that we would have introduced logic errors? The code was pretty straightforward, the algorithm wasn’t that unique or novel. It was probably new in some sense, but really just a collection of existing algos that were assembled in a new way. What makes you think it was such a hard problem to begin with?”

“Because humans are imperfect. I’m not even saying that there aren’t any logic errors in the first iteration -- err, the initial release, or whatever you want to call what I just wrapped up. What I am saying is that we’re building from a more structured base. We have good confidence that it works now due to the unit tests that we put in place. Sure, it’s slow, but we can always work to improve that speed; then we can regress over the unit tests to ensure that the underlying foundation is solid.”

The taller fellow shifted his feet on the hillside, lava rock crunching like broken glass under his boots. “I think those unit tests are a waste of time. Out of the two weeks you spent building the solution, you spent almost a full week on those. Some testing is great and all, but come on! You spent thousands of dollars on code our customer will never see!” He shifted again and this time a clatter of jagged rocks slid down the hillside and off the edge of a cliff about 30 feet below.

“Our customer doesn’t see the code for the unit tests, but it’s there for us to ensure that everything’s working. Tell me this, Bob: how would you have shown the customer that the code worked as they expected?”

“I would have let them watch it run -- watch it digest data and then view the results! That seems kind of obvious… how else would they accept it. Sometimes I don’t get you’re thinking, Fred”.

Fred kicked at the base of the column he was leaning against, flattening out a section of gritty soil, flecked with chunks of lichens that had fallen from the monument he now rested against.

“The tests are nothing more than the reflection of the requirements. The function that the customer described in their requirements is accomplished and realized by the collection of the algorithms and code segments that we built and assembled. If each of the blocks of code that we assembled works perfectly with regards to its own inputs and outputs, and if the code segments are assembled in a way that logically generates the expected output from the input to the entire routine, we can safely say that we’ve tested the whole thing when all unit tests are green. Sure, it won’t take into account the details of how much they like it overall, the architectural “goodness” of the solution, or the way it looks, the fonts, the UI, or even the speed, as you’re harping about. But if you remember, the customer never gave us any non-functional requirements like what I’m talking about. What they asked for was an algorithm and that’s what we built.”

Bob put his hands on his hips. “So you’re saying that the customer isn’t going to like what you delivered, and you anticipated that, but you still delivered it?”

Fred sighed. “No, but I am saying that the customer maybe doesn’t know exactly what they wanted when we started the project.”

“Oh, so now you’re smarter than the customer? Sounds like typical engineer ego.”

Fred squatted in the hollow that he had shovelled out with his foot and leaned back against the softer, mossy base of the rock, facing north. “I may be smarter or I may not be. That’s not the issue. The issue is that my job is not just to give the customer exactly what they say or to try to tease every possible requirement out of them at the beginning. My objective is to make a customer delighted with the software that I’ve written for them. It should meet their business needs and it should be useful, and it should be pleasant to use. That’s true of all software -- that’s true of all products. What my customer wants is to solve problems, and that’s why they agreed to pay me money.What I am saying is that right now, my customer has a piece of software in front of them that works. I haven’t said I’m done, I haven’t said that the software is perfect, I’ve merely presented them with a quick release and asked them to take a look and then talk with me about it.”

Bob dropped his arms from his sides and attempted to squat on the hillside, resulting in more lava rocks tumbling down into the ravine below. He finally settled with one foot pointed down the hill, and the other uncomfortably folded under him and pressed up against some unpleasantly sharp looking chunks of obsidian. “I just don’t get it. You want to talk with your customer about slow software that doesn’t look pretty and only runs in an ugly console. Don’t you think they’ll be pissed with what they see?”

“It depends. We talked before I started work on the project. I explained what I was planning to do. I explained that I write software in iterations. They may not be used to doing business this way, but I think that their expectations are exactly in line with what I’m giving them. They may doubt in the approach I’m taking, but I didn’t lie to them and I’m not delivering anything other than what I said I was planning to deliver. They know I’m a professional, and they expect me to do my job.”

Bob raised his eyebrows. “So you think they’ll be OK with this? Aren’t they just going to add a whole bunch of those requirements that you were talking about? Speed? How it looks? How easy it is to use?”

“Probably,” Fred shrugged. “It doesn’t really matter if they add those things now. Like I said, we have a solid base. If they want a different user interface, we’ll add that. If they want it faster, we’ll optimize it. What I’m happy about is that the code that I’ve just delivered actually will work. If they wanted to, they could start using it now.”

Bob started to raise a hand, but the whole swath of rocks below him began to slide. Fred grabbed a branch of a small and unhappy looking Douglas fir that had managed to sprout in the crack between the basalt column and the hillside, and hauled Bob up closer to him. “Watch your step there! That hillside isn’t stable at all!”

Bob muttered a thank you under his breath and sat down as close to Fred as he could comfortable position himself. Both men sat for a while, staring at across the hillside and down into the seemingly endless valleys and terrain beyond. The sun was beginning to set, and off to the west the sky was lit up with a brilliant gradient of reds, oranges, and pinks.

“Did you know that sunsets are pretty mostly because of the pollutants in the air?” Bob asked. “If it wasn’t for humans burning things, the sunsets wouldn’t be nearly as dramatic. I’ve heard that it’s the imperfections in the atmosphere that give sunsets like this so much colors and variance.”

Fred nodded. “I’ve heard that too -- not sure if it’s true, but it sounds reasonable.”

The men fell silent again.

“It’s like switchbacks” Fred said suddenly. Bob looked up, uncertainty on his face. “What I mean,” continued Fred, “is that the iterative process I’m describing is like switchbacks on a mountain. You don’t just charge up or down the mountain. It’s too steep,. the footing isn’t sure. Better to progress from east to west, then west to east, then back again as you slowly gain elevation like we did coming up. It’s slower, but it’s more steady. Plus, if you’re talking about a path that you’ve never been on before, or that hasn’t been blazed by someone in the past at all, the movement back and forth gives additional perspective. As you traverse back and forth you get a better idea of the mountain you’re ascending. You can see whether your path is doomed to failure or if you can realistically reach the summit. Each switchback can feel slow and painful but you’re safe and you’re constantly able to see what the next steps are going to be. You’re constantly analyzing and re-analyzing while still making progress towards the top.”

Bob looked down. “I guess I can see something in what you’re saying. So you’re saying that in some ways, software engineers are like sherpas, leading their customers to the summit?”

“I like that,” Fred smiled. “It’s a lot like that. We know that our customers want a great experience, and we’re the professionals that know about how to get there. We don’t want to promise them a spectacular view only to find that they were hoping to see Mt. Rainier and we we’re leading them up a slope to see across the Puget Sound. We could ask them a million questions about every last detail before they begin in a questionnaire. Things like ‘do you like flora more than fauna?’ or ‘do you prefer trees to rocks?’. But for a lot of the people that climb these hills, they wouldn’t even be sure what you mean. What types of flowers are there? What types of trees? They need to experience a part before they know what they want in full. Best to give them something good soon and then tailor the rest of the hike based off of how they react to what you’ve given them.”

Bob laughed. “I never really thought of it like that. I guess I can buy that. What about all that stuff you were talking about unit tests in code? Does that have a parallel in your analogy?”

“Well, I guess all analogies are somewhat imperfect, but I see things like tests in this case to be the safety that you build in. You may end up crossing a particularly treacherous avalanche chute, like this one, you almost slipped down. It’s better to spend a few minutes either picking a really good path and marking it with a cairn. Or maybe you’re crossing a creek and you spend the time to stretch a rope across to make it a bit easier to pass through. You expect in all likelihood you’ll be back on this path at some point and you want to be sure that the path is safe and trustworthy. Unit tests are like that in a way. It’s not really wasted time. You end up safer the first time around, and also, you have more confidence when you return.”

Bob sighed. “So you think the customer will be happy?”

Fred picked himself up off the edge of the dry, dusty basalt block and stretched. “I think so, I really do.” A crepuscular pika poked his head out of a hole about 20 feet away for a few seconds, saw the men, and decided to call it a night. The sun was nearly down now and the two men had a bit of a trek back to their base camp in the valley below. “You never know though, some customers are never happy. Some people will never be content. Ultimately, you can’t please everyone.”

Bob scrambled up, brushing dirt and shards of rocks from his pants. “I guess not.” he said.

Fred turned and starting walking back down the rocky path that they had come up. He paused, as they passed the first switchback on their trip down. “You know, Bob, even though hikes aren’t always as much fun as I thought they might be, and even though you end up with blisters and scratches and maybe the mosquitoes are biting, I still enjoy them.”

Bob just smiled.