vava’s ideas

If you have spent any time working as a software developer you probably know that the most essential skills in our field is the ability to make the right choices. No, really, every single time you writing a code you have to think, do you want to make all that iterations in a loop or use a recursion, do you want to use clever functional style techniques with closures and stuff or just OOP, or maybe just plain procedural code, do you want to store your data in a database or in a simple file and so on and so forth.

But there is a one choice you have to make even before you start your favorite editor and it is - should you write that thing yourself or maybe just reuse someone else’s code from out there?

It is really, really hard question. From developers point of view it always fun to create something by yourself, you’ll have fun and get more experienced. And it will do exactly what you want it to do. But from business perspective it’s easier to reuse or even buy someone else’s code, since it’s already written, bugs are fixed and you have guys outside of the company working on it so you could use them as help and they also will fix bugs in your code for free.

In most cases the choice is obvious, nobody would write his own database if he can use mysql, mssql or whatever and it’s enough for his needs, nobody in his mind would create his own language to use it internally, except really specific cases when that actually makes a point, like in FogBugz for example, nobody would write his own graphical editor if what he need is just to edit couple of dozens of photos. But you see, sometimes you have to write that stuff yourself even if it’s complicated and will take a lot of effort and time to build. The obvious rule of thumb here is to write only the core set of features, something that will distinguish your application from it’s competitors and download all the other parts from the Internet.

But there are situations when there is no right choice and one of them is when you want some tool to make some job just for yourself. Like you have to find something in a bunch of files, are you going to use grep or write a Perl script for that? What if you have to generate or edit some of the text files based on other text files, are you going to write sed script or a program in Ruby?

There’s no right or wrong choice here. Some of us will write ruby code faster than read man page for grep and find out what options he should use. Others will write sed script faster than figure out how to write in a file in C++. And the speed is not always a benefit for a choice, learning something new might be much more enjoyable and will pay off in a long run.

Anyway, the reason of this whole post is that I recently chose not to write and I want to tell you a story about it. And proof to myself the choice was correct.
Couple of weeks ago I bout an bluetooth headset and wanted to use it in a car with my Kaiser. But I’ve noticed that despite it is plugged in to car charger, simultaneous usage of GPS and Bluetooth drain battery in about 5 hours. Not usable. Then I remembered that some phones have a feature to turn bluetooth on when incoming call is arriving and disable it afterwards. I’ve checked that headset pairs automatically once bluetooth on a PDA becomes active so I just had to find a simple program to turn it on and off depends on call events. And I couldn’t find any. Just an empty spot here, really. But googling shows me MortScript for similar problem and I gave it a go. The first revision of a script was simple - it was waiting for a window with title ‘Phone - incoming call’ to became active, turn bluetooth on and then wait until all windows with ‘Phone’ in a tittle will go in to the background. A little bit later I added feature not to turn bluetooth off if it was on already.

while(TRUE)
        while(WndActive("Phone - incoming call...") = 0)
	   Sleep(1000)
	endwhile
	if ( RegRead("HKLM", "System\State\Hardware", "Bluetooth") = 8 )
           Run("\Program files\vijay555\VJVolubilis\VJVolubilis.exe", "-blueon")
	   while(WndActive("Phone"))
                Sleep(5000)
           endwhile
	   Run("\Program files\vijay555\VJVolubilis\VJVolubilis.exe", "-blueoff")
	endif
endwhile

You would have to install VJVolubilis and MortScript for that to work.

I’ve spent couple of hours on that script and it worked like a charm with one little detail, it was running a infinite loop eating up precious CPU cycles and draining battery. Sadly, MortScript doesn’t support events all that much.

But then I SKSchema came along, application that knows about events and could run some predefined actions on them. At first I just added two actions - turn bluetooth on incoming call and turn it off on disconnect. That worked except it would turn bluetooth off after call even if it was on long before call arrived.

So I removed a loop from MortScript and made SKSchema run it on call. Script then based on active window title will figure out when call is over and turn bluetooth off.

if ( RegRead("HKLM", "System\State\Hardware", "Bluetooth") = 8 )
   Run("\Program files\vijay555\VJVolubilis\VJVolubilis.exe", "-blueon")
   while(WndActive("Phone"))
        Sleep(5000)
   endwhile
   Run("\Program files\vijay555\VJVolubilis\VJVolubilis.exe", "-blueoff")
endif

And then I got thinking, as a developer, I could write that code in C# or even C++, it will work faster, it wouldn’t depend on windows titles and use normal events. It will be better.
But I already got the working solution. There’s no point in spending even a couple of days on that task anymore. Nobody except me will ever use it anyway.
So I decided leave it as it is. It automatically signed me off ‘Done and getting stuff smart’ category since those guys don’t think they just do but I knew I didn’t belong there anyway.

Huh, long post, all the right arguments, I know I am making a right decision here and still something bothering me. Yes, I just want to spend considerable amount of time learning Compact .NET framework, knowing I will never use this knowledge, and nobody except me would use the result product and I do have working solution already. Only my laziness helping me staying on the sane side right know. I guess it is damnation of developers, writing code even when you really shouldn’t. Please give me the slightest reason to do that, say that you would use this product and I will give it to you for free, no strings attached. Let the choice I want to win actually win. I am at your will here.

As a developer I often make various bugs in my code. It’s inherent in our job, we all do that, we’re humans after all. It’s no wonder that in some companies you even got asked about your favorite mistakes you made, if you didn’t make any means you didn’t write code enough. On the other hand it is a good practice to think about bugs you made, why that happened and how can you prevent them in the future. You’ll become a substantially better programmer after any single mistake you made if only you investigate it afterwards. So don’t be afraid of making one, you still will at some point of time, it’s inevitable in process of becoming better.

I definitely not afraid of my errors that’s why I decided to share them with you. Those a recent ones, I made them in a course of last couple of weeks of writing C# code and I novice in .NET, so enjoy.

TimeSpan.Milliseconds

This one isn’t really my fault. It’s Intellisense’s one and style of programming it imposes on. But let’s see from the beginning.

So one sunny morning I was tasked to make some research on how fast application would run in multi-threaded environment given that one crucial resource should work only in single-threaded manner. So I made a little test application, span couple of threads and time function calls I want to have results for. How do you usually do that? Of course, call Date.now before and after the call and then subtract former from the latter. Usually (that is, what I’ve got on every platform I have ever worked before) what you get after that is a number of milliseconds call took. But in .NET for some reason you get an TimeSpan object. After a couple of seconds of frustration I called an Intellisense to see if there is a method to get number of milliseconds this object is represent. So I typed a letter “m”, and I saw a really nice property named “Milliseconds” there. “Ok”, I thought, “looks like the one” and I put it there. After a couple of tests with different number of threads I noticed something suspicious. Whatever the number of threads was, the average time always stays around 500ms and max time was always 984ms. At first I thought that is a good news, that means application is good written and we’re not going to have any performance problems whatsoever. I even told that to management and they were happy. But my intuition was telling me that something is wrong here. Calculations of average time was taking so long given that average is half a second. I mean common, function was called 12500 times but it would take only about 1.7 hours to run and it took like 4 of them. Something was really wrong.

There were couple of additional locks in the code that were for collecting statistics and so weren’t timed so I decided to eliminate them and I did. Then after every operation I put a code that writes a “.” in a console to see how fast a progress goes. And even though there weren’t any untimed locks and average time stays half a second and max time 984ms still I could really see sometimes progress stops for about 5 seconds doing nothing. That’s why I decided to check my timing code. Magical 984ms was obvious error. After couple of minutes staring at the simplest code I ever wrote I decided to go check what MSDN knows about TimeSpan. And there it was, somewhere in the end of an article about TimeSpan.Milliseconds was a remark that it could take a values from -999 to 999 only, and was for a fact just a component of interval whole length. That’s why average stays 500, just in the middle of possible values (negative values weren’t possible in my situation). No link to correct method though, quality of MSDN is really low these days. But luckily there was a sample code right in the article with calls to all possible TimeSpan functions and results they return. That’s how I found the right property for a job, TotalMilliseconds. And yes, correct performance values were never near the 500ms, more like 40s. Sad news but a different story.

new Random()

This is a really funny story. The most unexpected error I ever got with random numbers. Here it goes.
As a part of fighting with TimeSpan.Milliseconds bug I decided to run all my threads at the same time. Just to make sure none of them finishes before the full set of them run. At the same time I used a lot of random stuff around, to choose values for function calls, to create a users (I was actually test client to some kind of special purpose database, not a fancy one though), to choose what to do with database (that is, create something or just get data from it). After the session each thread was deleting users it created to clean a DB for the next round. And I’ve started to get strange results here, while deleting a user DB sometimes would complain it doesn’t have that one. Hm. It was created before after all, where did it go. It was late and I was tired from Milliseconds story already so I just ignored that complains and removed a code that checks it. Which leads to even more strange results, sometimes in a function user was created, uploaded to DB, everything was fine and then just a couple lines later it couldn’t be found. Not a nice stuff.

The next day I fired up a debugger to see why that happened. After couple of runs when I couldn’t find out a reason for such a strange behavior, I decided to put checks in cleaning code back since those issues probably was related. And when the check failed and I was looking at what other threads doing I found out couple of them were cleaning as well. Sadly, VS 2008 couldn’t show a collections while thread is paused so I couldn’t look up what they were cleaning but luckily one of them have value cached in local variable. And you know what? There was the same value in a collection of a thread I was examining.

So random were broken. How could that happen? And then a lucky guess, could it be that… I put a breakpoint just after a first random value was generated and, you guess it, for every thread it was the same. Yeah, random get seeded by the time, apparently it uses not high resolution clocks and threads which runs simultaneously even on single core computer and under .NET VM could get the same seed. Really nice. Adding Thread.CurrentThread.ManagedThreadId to the (int)DateTime.Now.Ticks and seed randoms with this value cured the bug instantly. This going to be my favorite bug with Random values for a long time, I don’t really know what could beat it.

System.Threading.Timer vs. System.Windows.Forms.Timer

This is the bug from a different application. This one shows how bad MSDN is.

So I was making application that should every couple of minutes do some work and notify user after it’s done. Application was working silently in the background so as a notify facility I chose TaskbarNotifier, it does a great job of informing user without disturbing her.

So, how do you run a task every couple of minutes? You going to use timers of course. What the first thing you see if you look for “timer” in MSDN? System.Threading.Timer, which is, and I cite, “a simple, lightweight timer that uses callback methods and is served by threadpool threads. You might also consider System.Windows.Forms.Timer for use with Windows forms, and System.Timers.Timer for server-based timer functionality. These timers use events and have additional features.” Apart from threadpool reference is what I wanted, simple and lightweight. So I ignored that sign and go coding. Implementing was easy, everything worked just fine until I finally got to the point where I wanted to notify user. Notification just wasn’t working at all.

I saw a demo and knew it should work so I start debugging. Got into library code and noticed that timer callback that lib uses to slide window slowly just doesn’t fire. I found this thread but I thought it doesn’t apply to my situation, I do all my stuff in a single thread. After minutes of frustration and googling I decided to rewrite TaskbarNotifier a little with System.Threading.Timer since I knew it work well. And I did, and it crashed with the exception saying some GUI control was used in a thread different from where it was created. And then it hit me, “Threading” is all about threads, isn’t it? And I read documentation carefully and yes, this timer spans a new thread for callback. Not sure how this is “lightweight” but here you go. Yes, changing System.Threading.Timer to System.Windows.Forms.Timer made everything work just fine. But hell, this line about new thread should be right there, in one-line description of a class, that’s what I read after all. And yes, it’s not near simple or lightweight, so it shouldn’t be stated there. And yes, the first class in search should be most used one, System.Windows.Forms.Timer, what’s wrong with relevancy there?

That’s it. I didn’t make any more mistakes those couple of weeks. I did much more in my life of course but well, as I already said it’s inevitable. Be proud of your mistakes, they make you better.