Kevin Jenkins' Blog

I spent about a week integrating Angel Script and exposing the Ogre function calls. My graphics programmer spend 3-4 days doing the same. In the end all that got done was billboards from scripts. This is because we had to implement so many work arounds for bugs and missing features in Angel Script that it was faster and easier just to do everything in C++, even not counting the initial setup time. For example, there is no way to have per-effect variables in script, and you can’t return pointers to objects that have member functions. There was no way to know this until really trying to use it, so I don’t place too much blame on myself. It was just bad luck. With more time invested these problems could be circumvented, but considering we already spent 10 days doing something that could have been done in 5 minutes in C++ it’s not worth the additional cost.

Most likely all further graphics work will be done in C++ and this will be revisited after the game ships.

Here’s some possibly lesser known tips to move fast while programming (C++, Visual Studio)

Windows

1. Add explorer to the quick launch bar. Under target, use “%SystemRoot%\explorer.scf”. Under start in, use “%HOMEDRIVE%%HOMEPATH%”. Both without the quotation marks. This way you can get to C: with one click, rather than starting at My Documents
2. Similarly, in explorer, you can add local directories to the favorites list. I added the base directory to my game there.
3. If you don’t have the show desktop button in the quick launch bar, add it.

Visual Studio

1. Hit ctrl-tab and hold down tab to go between the most recent windows
2. ctrl-minus to go back to the last cursor position (can be done repeatedly). ctrl-shift-minus to go forward
3. ctrl-shift-v to get a list of recently copied items.
4. Memorize and use the hotkeys for find in files.

Visual Assist

1. Use alt-g to goto on a keyword or token. If the results are wrong, right click, go to definition or declaration. Between the two you can get to where you need to be.
2. Use VAssistX under the menu bar, then go to refactor. Renaming there is very useful. You can also get this by getting a tooltip for a variable or function, then clicking the down pointing black arrow that usually comes up.
3. Shift-alt-o to bring up a list of files, searchable by name.

UltraMon and multiple monitors

1. Under options, select Smart Taskbar, use Smart Taskbar, and Mode: Standard. This gives you two task bars.
2. Under options / Hotkeys / Move window to next monitor, I’ve bound alt+n. This way I can view 2 fullscreen documents without reaching for the mouse
3. Both of my monitors run at 1600×1200
4. When multiple copy/paste isn’t enough, I open UltraEdit as a notepad and can move quickly. UltraEdit also has a more powerful find/replace tool than Visual Studio and is better for dealing with multiple line replaces.

So I spent the last 5 days week implementing and exposing all of Ogre 3D to AngelScript, and I’m starting to feel it was a mistake. All I’ve gained is the ability to run code in text files, and reload them at run time. But that benefit is very limited – slow, hard to debug, less capable than the compiler. And for what I’m using it for I could have programmed it in C++ directly in 15 minutes. As a long-term solution, like if I had a year to develop this, this is a good solution because these text files could serve as data files as well and be output and worked with from an editor, with the ability to debug the script itself. But as a short term solution I’m feeling I wasted 5 days. I have nothing visible to show for my work and it’s not like I have plenty of time to spare. And the worst part is I’m not sure it’s going to work. Angel Script is pretty hard to use and understand. The forum support is good, and what they are doing is a difficult task so understandable to an extent. However, me figuring out means stuff takes twice as long as it should. For example, how do I handle casting references to pointers, or vice-versa? Why does declaring a parameter list of (void) not work, but () does work? Why did they decide not to support pointers, but to require reference counted objects? This means not only do I have to declare script classes for all my game classes, but I have to do it duplicate again for the pointer to that class as well.

If I could do this week over again I should have just hardcoded everything.

I had a guy from India call me today. He runs a programming company there that offers game related outsourcing for $25 an hour and offered to take on any tasks I might have.

I probably shouldn’t have said this, but the first words out of my mouth were “That sounds good, but I have very high standards about who I will work with, and I don’t work with junior guys.” I then went off on a short rant about how I hired guys from India before and everything got screwed up due to mindless literalism. I gave the example of how I asked for shader light support in my game, so my Indian ex-programmer implemented a shader that only supports one non-directional light ever. And although I said it in a nicer way, I ended with “I don’t know if all Indians are like this or not but if they are I don’t think your guys can do the job.”

Now that I reread what I wrote, it comes off as a lot meaner than how I said it, but the essence is correct. I’m just sensitive right now because I was really screwed by that whole ordeal. I lost 6 months of graphic development and over $10,000 paying the guy, not a line of code he wrote is still used, and now at the very last minute before shipping I have to hire someone really expensive to salvage the situation.

Anyway, we agreed they can start with a smaller task and see how it goes.

I sent them the spec to implement the Autopatcher GUI interface for Windows, for RakNet. This one got screwed up by mindless literalism about 6 months ago. This was back when my ex-programmer from the Ukraine was working for me. While he literally did design a GUI interface, it was the simplest thing you could do and still call it a GUI. It totally ignored the intent of the task, which was to make a nice looking interface with a graphical background, buttons, etc. that the artists could fill out. I was actually better off with the original console window than what he did, and it wouldn’t have costed me 3 days of pay and development time for nothing. On top of that he used some proprietary libraries with incompatible licenses with RakNet without asking first. Even if he had done something acceptable, I wouldn’t have been able to use his work.

I actually said in an interview a couple of days ago “I don’t like coming to people with problems, I like them coming to me with solutions.” That’s why my programmer from Portugal is so good. He thinks about the intent of the task (to make a good game) and tries to do a good job, rather than the bare minimum. So I can be confident that the final solution is smart and usable. It really helps having someone I know can do the job, and moving the schedule forward.

Somewhat related, so far I like outsourcing so far more than hiring people, except for programming:

1. I don’t have to spend time recruiting and interviewing
2. If they screw up, it’s their time and money loss, not mine
3. The pressure is on them to meet the agreed deadline, rather than on me.
4. It’s actually cheaper when you consider that I don’t have to train people, I can hire around the world, and I don’t end up with the situation of paying someone for days/weeks/months with nothing to show for it.

It’s a good way to go for art and sound.

My first attempt at an effects system was to assign it to my graphics programmer, who decided to define effects in an XML file. The theory was that you could do something like

[Effect]
– [Particle “fire”]
— [Play for 5 seconds]
– [Particle “smoke”]
— [Play for 2 seconds, delay start for 1 second]

This didn’t work out, in part because the implementation was mindlessly literal so didn’t solve the fundamental problem, and in part because of a lack of flexibility to begin with. As the system became more complex, more and more parameters have to be exposed and you just can’t do certain things.

The second approach, which I came up with, was to tag 3D models with triggers and effects based on locators. So you would place a locator in a file and define in the object properties an effect and a trigger. For example, on the thruster of the fighter I would have “Trigger=ForwardThrust Effect=ThrustParticle” This works, and is good for placing simple location based effects, but is not a complete solution and is hard to use and inflexible.

The fundamental problem, which I didn’t consider sooner, is that effect is such an open-ended term. It would be like if I asked you to write an effects system for a magic based RPG. No matter what system you came up with, I could come up with a reasonable counter-example as to why your system isn’t flexible enough. Effects (especially magic) do everything from lights that spin around the character, to glowing a specific part of the character, to time based events, to modifying and creating meshes, to reading specific game properties. You can’t really apply constraints to a fundamentally unconstrained system. Essentially you see stuff, and stuff happens, based on other stuff.

My third, and I think final approach, is to implement effects with scripts. So I’ll specify triggers in an XML. Each trigger references a script. And the script does all the effects, whatever that may be. If I need to tag specific points on a model, I will just lookup the locator in the .effect file. This should give me the flexibility I need, while still being easy to use and work with.

It’s one of those things that sound obvious when you hear it but wasn’t so obvious to begin with.

I wrote about AngelScript a while back yet never had a reason to put it in the code up until today. This morning I was thinking about how to go about a general effects system for my game and concluded that the only thing flexible enough was either code or script. I specify effect files in XML settings, and those can’t refer to code (easily) so that left me with script. Plus this is really one of those rare times when scripting makes good sense:

1. Used in a limited fashion that won’t require much debugging
2. Extensible by modders in a situation where I don’t want to give out the source code
3. No better solution. I could make a full-fledged editor if I had the time and money, but I don’t.

I started on this about noon, went out for a good chunk of the day, and finished integrating it into the game now. I also created a sort of script instance system, where a script module (one or more related script files) is treated as a mini-program, with a per-tick update function, and a query function to indicate when the program was done. The script manager automatically handles tracking the pointers, and I did a pretty good job with the architecture the first time through for once, partly because my wife made me get the laundry and that 5 minutes away from the computer helped me solve problems in my head before I spent a lot of time coding. So this instance system will be perfect for effects as they update every tick doing stuff, and end after some amount of time.

I think it’s pretty good work that in one day I went from the vague problem of how to implement effects, to a working script framework for it.

I rejected 4 job candidates in as many days, using the new technique of “Give them an assignment, see what happens.” The difficult question right now is what kind of assignment I should give. I sometimes feel like what I’m giving is too hard for a junior and if I had interviewed myself out of college I wouldn’t have hired myself – a mistake I should avoid.

3 of the techniques I tried are:

1. Look at this shader code I wrote and give me feedback on how to optimize or improve it
2. This lookup is using a linear search. Change the data structure to an optimized list.
3. Write code to send a list to the server, have the server sort it, and send it back

The answers I’ve gotten are:
1. Uh, it looks OK. (On further code review the guy didn’t understand basic C++)
2. Implemented using a map, instead of an ordered list, and was incorrect as he didn’t understand how to use C style strings
3. Had the client sort the list instead of the server, and attempted to write quicksort on the spot rather than using the existing ordered list class I provide, or stl, and also couldn’t make a project for RakNet after 1.5 hours even though I give example projects and instructions on it.

So far there’s only one guy I rejected out of hand due to price, which was $35 an hour where I knew I couldn’t talk him down. He was also the best candidate by an order of magnitude, with real experience. I’m on the edge about that one. I can’t afford it, but I feel someone like that could make real progress in the game.

I’ve been trying to contract out sound effects. I have over 200 sound effects for my game, most of which are repeated variations. For example, for the cloaked mine weapon I have

CloakedMineActivate.wav
CloakedMineBulletAmbientSound.wav
CloakedMineBulletHitAsteroidSound.wav
CloakedMineBulletHitDestroyableSound.wav
CloakedMineBulletHitFighterSound.wav
CloakedMineBulletHitWallSound.wav
CloakedMineFailOnAmmo.wav
CloakedMineFailOnEnergy.wav
CloakedMineShrapnelAmbientSound.wav
CloakedMineShrapnelHitAsteroidSound.wav
CloakedMineShrapnelHitDestroyableSound.wav
CloakedMineShrapnelHitFighterSound.wav
CloakedMineShrapnelHitWallSound.wav

For the 20 different items and weapons I have I got a quote back of $17,000, which is about $12,000 more than I was expecting. So at this point I can either cut things down to some important component (enough to ship), keep looking for another company, or only have them do critical ones, and do the remainder myself.

I didn’t expect sound to be this expensive – I guess it’s another lesson I should have learned before I started.

Today in my interview I didn’t even bother with the usual process of asking technical questions. All I did was say “Build the game” and after that was done “Implement this task” Building the game took an average amount of time, but the task was a 10-15 minute task he took an hour to do, and did wrong, so I ended the interview and there were no doubts from either of us that I made the right decision.

Total time taken away from me was probably 15-20 minutes, and I got a perfect measure of the candidates suitability.

When I think back on my previous bad hires, this approach would have screened out all but one of them, and it would have even worked there if I had given a more complex task.

It seems obvious now, but I can see how my past approach of asking lots of technical questions wasn’t sufficient in itself. All that did was keep away guys that didn’t know the performance differences between a linked list and a binary search tree. Since most candidates do, this didn’t help me that much.

The best approach I have right now is:

1. Prescreen with technical questions that require good fundamentals (use of data structures and 3D math). This should only take the minute or so to send an email, and another 5 minutes to scan over the reply. Also require salary requirements and start availability in the reply.

2. Give a short but real task, that involves working with other code, that is typical of the kind of tasks you might actually give. This will tell you if the candidate looks at what he is doing carefully, can read and work with existing code, can follow and understand instructions, and how fast he works. I think speed is a good measure of ability.

3. Hire on an hourly basis at first, with a clear understanding that the first few months are probation. This will save you from guys that are lazy, or who got past step 2 and shouldn’t have.

I’m happy that I saved a lot of time and money today with this new technique. Too bad I didn’t know it months ago 🙁

I got another bad hire a couple of days ago. One thing I noticed is that even though I spent 2 hours carefully interviewing him, it only took me 5 minutes of working with him before I started seeing red flags and warning signs. Among other things it took him 5 hours to compile, download, and run the game, when I’ve had others in the past do it in 30 minutes. That’s raw experience right there.

I’m going to try a new interviewing technique, which is to give a real programming task that

1. Interfaces with a larger system, so you have to work with someone else’s code
2. Has multiple solutions, some of which are better than others
3. Takes enough time to require real though and design, but not so much that it would take someone competent longer than an hour, and designed such that a great person could do it in minutes.

I’m trying to think of some good problems. Right now I’m partial towards giving an assignment to extend RakNet in some way. I have a guy scheduled for an interview tomorrow and will see how this goes.