April 29th, 2015
I’ve spent four days at the beginning of March this year working on immersive hero images in our app. I touched around 250 lines of code all in all, mostly adding new code but also modifying or deleting a few dozen existing lines. And that’s pretty much all I did in those four days. Sure, there were a couple of small meetings here and there and an occasional code review, but overall I’d say that I’ve spent at least 8 hours every day working on this feature. Overall my total output was less than eight lines of code per hour. That sounds not very productive of me, I know.
Recently I find myself not really writing a lot of code. Instead of diving straight into the editor, I think. And I think. And I think some more. I literally sit there staring at my desk. I get up and walk around the room. I take a break and browse the web (but don’t tell my managers about that). And then I push it out to the next morning.
On any given day I find myself spending, perhaps, 10% of the time writing code, 40% of the time debugging it and 50% of the time thinking about it. It certainly seems like a big waste of my time. I should be down there in the editor, banging out line after line, method after method, class after class, throwing it at the device and seeing what’s happening. You know, walking the walk.
But I’ve walked that walk before. It felt good to always be moving and it felt great to always be moving fast. These days I prefer to move slow.
I like to think about this as deliberate brevity. A long long time ago, a French mathematician and philosopher Blaise Pascal said “Je n’ai fait celle-ci plus longue que parce que je n’ai pas eu le loisir de la faire plus courte.” Translated loosely, “I have made this longer than usual because I have not had time to make it shorter.”
I enjoy taking time to make my changes brief and vigorous. In the words of William Strunk, “When a sentence is made stronger, it usually becomes shorter. Thus, brevity is a by-product of vigor.” I believe that every line of code needs to fight for its right to exist in our code base. I believe that every comment needs to fight for that right as well. I immensely enjoy code reviews where one of my teammates points out how I can cut out even more code and make things even simpler, but without losing the overall clarity of what the code is doing.
I am indifferent to productivity tools in general and to debates on which IDE makes programmers more “productive”. When you spend five times as much time thinking about the code than writing the code, the number of shortcuts, templates and auto-completion becomes quite irrelevant. I don’t measure my output by words of code per second or lines of code per minute. I measure my output by the robustness of the path I’m paving, over weeks, months and years.
When you spend half your day thinking about the code, every distraction hurts. Open office plans that proclaim to foster spontaneous exchange of ideas are creating a never-ceasing cacophony of distractions that is actively disrupting the most precious resource we have to offer our employers – our brains. Until they come with a thinking-time switch that envelopes you in a literal cone of silence, they must not progress beyond the corporate facilities planning spreadsheets. I realize that it’s a bit late for that.
Don’t aim to impress me with how fast you can churn out new code. Don’t tell me that it will all be done by the end of business day. It rarely happens, and when it does, I know I’ll find myself wading through vast reams of that code in a year’s time (when you might have already gone to some other project), trying to understand why it is, in fact, so vast. Don’t ask for permission to spend time thinking. And don’t think that spending all that time thinking only to write fewer lines of code makes you a less productive programmer. Be concise. Be brief. Be vigorous.
April 22nd, 2015
In the epilogue of “The Mythical Man Month” Frederick Phillips Brooks says that there are only a few people who have the privilege to provide for their families doing what they would gladly be pursuing free, for passion. I am quite lucky to consider myself to be among those people. In a hypothetical situation where I buy a lottery ticket and win the 8-digit prize, I’d most probably find myself not necessarily continuing programming each day, every day. Having said that, it’s quite nice to be employed doing what you love to do. But, as with all things in life, you can’t expect a steady level of “enjoyment”.
Years and years ago (I’m talking last century) when I was studying computer science at a university, one of the mandatory courses included a homework exercise doing a stopwatch in assembly. You had to write a program that gets user input in mm:ss format, and when the user hits Enter, starts counting down to zero. I think there was also the part where you had to track keyboard input and stop the timer, but my memory fails me at the precise details.
I hated that exercise. I absolutely abhorred the very notion of dropping so low to the metal. You know, pixels are my people. So I waited. And waited. And waited. Until the very last evening before the morning when the homework was due. And then I schlepped to the computer lab, plopped into the chair and considered the bleak night in front of me.
As a parent I tell my kids to not postpone the exercises and chores they don’t like until the very end. I tell them that if they start with things they enjoy, they end up with a clump of boring, tedious and dull things at the very end. Of course back then I didn’t want to be quite as reasonable. So as I forced myself to start with something, anything, anything at all, I picked up the less mind-numbing parts of the exercise.
I started with processing the user input. Something like look at the character that was just typed and beep if it’s not a digit. Then proceed to mirror the character on the screen and advance the cursor. Then, at least after one digit, also accept the colon. Then tweak the input validation to only accept digits from 0 to 5 in the next slot. Then tweak the input validation to only accept Enter after exactly two digits after the colon have been provided.
And then I stalled for a bit more time, tweaking the input processing routine to accept additional Enter strokes to treat them as decrementing one second from the timer. Which brought me to writing another routine that would update the currently displayed value, handling the transition from :00 to :59 state.
I distinctly remember the depression I felt at the very thought of tracking and syncing with the system time facilities. I tried to postpone it as much as I could. It was the farthest away from the pixels, and I left it until the very end. And then a wonderful thing happened.
As I kept peeling off those onion layers, I kept on getting a continuous stream of visual progress confirmation. My routine for accepting the initial timer value was working. My routine for rejecting invalid inputs was working. My routine for decrementing the current timer value was working. And as I kept peeling off those onion layers, and as I got the the very last piece, I found out that it was actually quite manageable. There was only this last thing to cross off the list. I did it and I didn’t even notice how smoothly it all went.
Fast forward to early March 2015 and the code base I’m working on. It’s been around for a while. I’ve been on it for a while, almost five and a half years. It’s seen some major redesigns. In fact, I can’t think of a single module or class in it that hasn’t been gradually (and completely) rewritten at least once. And it has a lot of baggage.
I’ve briefly mentioned this the last time Chet and Tor hosted me on their podcast. As we started down the path of bringing our app into the Material world, we had two choices. We could graft the various aspects of Material (keyline grid, in-screen animations, cross-screen transitions) on top of the code base that was, frankly, simply not built to be that flexible. Or we could rebuild the foundation (ahem, yet again) with an eye towards that flexibility and then bring those elements in.
Rebuilding the foundation is at most times a long, dull and unexciting process. You know that at some point in the hazy future when it’s all done, it will enable all these wonderful things (and you hope that by the time you get to that hazy future, the design hasn’t changed in any major way). But for now, and perhaps for a while, the things you’re doing are not resulting in any kind of user-facing improvements. If anything, the deeper you dig, the worse it gets for the overall instability of your code base (aka Things You Should Never Do – except when you really have to). And so you dig, and you dig, and you dig. And you get closer to that hazy future. One step at a time.
And then at some point you look at what you’ve built, and you see that it’s ready. And you make a 150-line code change and suddenly you have the hero image extending all the way into the status bar on the details page. And the next day you make a 40-line code change that builds on top of that and you have the same functionality on the profile page – because they share the same skeleton structure that you and that other guy on your team has just spent the last four months rebuilding from scratch. Oh, and there were these ten people scattered all across these other teams that built the components that enabled to rebuild that skeleton structure in only four months. But I digress.
And then the next day you have a 50-line code change that brings the same functionality to a more complex context that looks exactly like those other two you already did, but instead of a simple RecyclerView it’s a RecyclerView in a ViewPager that has a single tab which hides the tab strip where you don’t even know if you’ll need to display the hero image until after the second data loading phase that happens when you’ve already configured the initial contents of that ViewPager. But you didn’t start straight away from this complex case. You started from a simpler case, hammering out the details of how to even get into the status bar. And when you got to the ViewPager context, you were able to concentrate on only that one remaining case.
And so you go. Across the landscape of your code base. Sometimes it just flows and you don’t even notice how yet another week flew by. And sometimes you look at this craggy cliff and you just want to turn and run away. But instead, you work at it. One step at a time. And before you know it, the cliff is behind you. And then you take the next step. Because the next cliff is waiting.
April 15th, 2015
I’m not going to take credit for the story, nor would I claim a perfect analogy. But here goes.
Imagine you’re in a room with ten screaming babies. As all of them are screaming at the top of their lungs, you start feeding them one by one. You’re done with one, and there are nine screaming babies left. Some time passes and you’re done with another one, and there are eight screaming babies left. Some time passes and you’re done with another one, and there are seven screaming babies left. And it doesn’t feel like you’re making any kind of progress because as long as there is at least one baby screaming, you can’t get any peace.
That’s how it can feel to be a programmer on any reasonably sized project. Where screaming babies are bugs in your incoming queue. They never stop demanding your attention, and they never stop screaming at you. Unless you make peace.
If there’s one axiom of software development that I hold inviolable, it’s that there will always be bugs. You can surround yourself with a bunch of processes, or do extensive code reviews and endless testing rounds. But the bugs will always be there. If anybody tells you that their code doesn’t have bugs, just shrug and walk away. They have no idea what they’re talking about.
Make peace with the simple fact that the code you’re shipping today has bugs.
Some bugs are scary. You need to tackle them. Some bugs, on the other hand, are these little tiny things that simply don’t matter. The problem with most (probably all but I haven’t tried them all) bug trackers is that the scary bugs in your queue look exactly the same as the tiny bugs. Most of the time the only difference is going to be in the single digit in the priority column. Or maybe the scary bugs will have light red background across the entire row. Or maybe the tiny bugs will use lighter text color. But they probably won’t.
And so you stare at your queue and you feel that you just can’t win. No matter how much effort you throw at that queue, as long as you’re not at zero bugs, they are screaming at you. And every time you fix a bug, you touch your code base. That’s another bug that you’ve just added. And every time you fix a bug, you get assigned a couple more.
Make peace that not all bugs are created equal.
Zero Bug Bounce is a fiction that some people invented to make peace for themselves and to create an illusion that they are in control. So at some point in the cycle everybody looks at the pages upon pages of bugs in your project queue, frowns and then mass-migrates a bunch of bugs to the next release. And to the next one. And to the next one. And at some point some bugs have been bumped out so many times that you might as well ask yourself some very simple questions. Do those bugs matter? Do they deserve to be in the queue at all?
Make peace that not all bugs are actually bugs.
Sometimes a feature that you’ve added to your product just doesn’t work out. It doesn’t get the traction you’ve expected. Or it’s not playing well with some other features that you’ve added afterwards. Or you’re not even sure how much traction it’s getting because you forgot to add logging, and there’s nobody on the team who actually cares about this feature after the guy who did it left the team and you’re in the middle of the big redesign of the entire app and why should you even be bothered spending extra time on that feature. Phew, that was a bit too specific.
And of course there will always be somebody who used that feature. And now that you’ve taken that toy away from them, they are screaming at the top of their lungs. And you cave in and bring that feature back. Well, in theory at least. But it’s been redesigned to fit into the new visual language of the platform. And now somebody else is screaming at you because you’ve changed things. All they want is just a teeny tiny switch in the settings that leaves things they way they used to be. Sure, they want new features, as long as they look exactly like the old features. But that’s a topic for another day.
Make peace that your work is never done. That if you want your work to be seen, you have to ship. Make peace that the work you ship will have bugs. Take pride in things that work. Develop a sense to know scary bugs from fluff. And develop a thick skin to ignore the screaming.
February 24th, 2014
If only they used code guidelines that mandated braces around all blocks. If only they had unit test for this module. If only they had better static analysis tools. If only they had better code review policies.
There’s a lot of hand waving going around in the last couple of days, with everybody smugly asserting (or at least implying to assert) that they would never, in a million years, have made such a stupid mistake. And that’s what it is. Plain and simple. A stupid mistake. With very serious implications that reach into hundreds of millions of devices.
Except that stupid mistakes happen. To everybody. Unless you don’t write code. And if you write code and you really truly believe that you are not capable or making a mistake such as this… Boy, do I have a bridge to sell you.
Which brings me to my (almost) favorite thing in the world. Smugly asserting that I knew better than them and quoting myself:
My own personal take on this is that interacting with computers is too damn hard. Even given that I write software for a living. Computers are just too unforgiving. Too unforgiving when they are given “improperly” formatted input. And way too unforgiving when they are given properly formatted input which leads to an unintentionally destructive output. The way I’d like to see that change is to have it be more “human” on both sides. Understand both the imperfections of the way human beings form their thoughts and intent, and the potential consequences of that intent.
Do I have a solution for this issue? Are you f#$%^ng kidding me? Of course I don’t. But it kills me to realize that after all these decades we are still living in a stone age of human-computer interaction. An age when we have to be incredibly precise in how we’re supposed to tell the computers what to do, and yet to have such incredibly primitive tools that do not protect us from our own human frailty.