Category: Linux Game Development

Last time, I talked about Linux-specific issues to fix before my game’s release.

This time, I’ll address the issues I’m seeing on Android and Windows platforms.

Android: manual code signing

Quite frankly, between running the game on my phone and on my tablet, I haven’t seen any issues since I first tried to get my game built and installed on this platform. The main issue I had was figuring out which directory to save to, and I solved that issue.

Oh, and code signing was another solved issue. I can build and deploy debug builds by turning on developer mode on my devices, but the release build needed to be signed. As I am not using amazing IDEs that have one-touch buttons that do all sorts of fanciness, I had to figure it out myself from the documentation.

Luckily, the Android developer documentation for signing manually was fairly straightforward in this regard. In my CMakeLists.txt, I added a custom target called sign, which requires the location of my keystore and its alias. I created a few environment variables that I pass into my build, and the following is basically what’s needed as per the documentation:

ADD_CUSTOM_TARGET(sign
"echo" "================ SIGNING WITH KEY ================="
COMMAND "jarsigner" "-verbose" "-sigalg" "SHA1withRSA" "-digestalg" "SHA1" "-keystore" "${GB_KEYSTORE}" "bin/${ANDROID_APP_NAME}-release-unsigned.apk" "${GB_KEYSTORE_ALIAS}"
COMMAND "echo" "================ VERIFYING WITH JARSIGNER ================="
COMMAND "jarsigner" "-verify" "-verbose" "-certs" "bin/${ANDROID_APP_NAME}-release-unsigned.apk"
COMMAND "echo" "================ USING ZIPALIGN ================="
COMMAND "zipalign" "-v" "4" "bin/${ANDROID_APP_NAME}-release-unsigned.apk" "${ANDROID_APP_NAME}-release.apk"
WORKING_DIRECTORY "${PROJECT_BINARY_DIR}/android-project/")


Otherwise, I found porting to Android be very straightforward thanks to using the NDK and libSDL2-based libraries. If anything, I worry about scaling to different screen resolutions and device-specific compatibility problems due to the lack of devices I have to test on.

I’ve already signed up for the Google Play developer program, so the main piece to worry about is actually submitting my app to their store. How hard could it be?

Windows: persistence and font rendering

While GNU/Linux and Android are more or less the same, Windows is the odd duck.

I can easily cross-compile to create a Win32 build, and with my limited testing I found that the 32-bit version runs smoothly on a 64-bit system, so that’s good.

Since I don’t need to use a lot of memory, there’s no real advantage I can see to building a 64-bit version of my Windows port. The main downside would be an inability to support people on 32-bit systems, requiring that I provide both 32-bit and 64-bit binaries as I might need to do for the Linux-based package.

However, I did have to fix a few issues this past week that I didn’t know were there until someone tested it for me. Thanks, Rick!

I knew of an issue with using MinGW to cross-compile to Windows in which using std::cout would result in a crash. I never looked too hard into it because I only used cout for my own logging in order to find out what is happening, so I just commented them out when I released for Windows, usually for a Ludum Dare game.

Well, it turns out that there was still a crash, and I found that if I commented out the code that saved the current game state to a file, it would run just fine.

Was the known issue applicable to file stream operations, too? Luckily, gdb can be downloaded and run as a standalone applications on Windows, so I ran my game on Windows through gdb and read through the stack trace. It pointed to yaml-cpp.

I use yaml-cpp to save and load my game data, and it works very well. But why does it crash on Windows?

I found this thread on GitHub that mentioned a similar stack trace: Crashy shared object

It was closed without really being addressed, as the original poster gave up after seeing the issue disappear when using a later version of gcc.

Luckily, someone else found a different solution involving a change to a few lines in yaml-cpp’s code, although they said more tests are needed. I tried it, and it seemed to solve the problem for me, although I am a bit wary about not knowing what the change does or how it solves it. B-(

The other issue I found on Windows was that resizing the window results in the text looking completely wrong:

All the other graphics look fine. Under the hood I am using SDL2_ttf, but using it directly isn’t showing this problem. I am using NFont, which does some caching, and I wonder if it is somehow being corrupted. I need to do some more tests, but this issue does not occur on my Ubuntu system, and Android doesn’t allow you to resize the screen dynamically at runtime, so it’s a Windows-specific issue so far.

I’ll continue looking into it, but updating to the latest version of NFont didn’t help. I tried updating my SDL2-related libaries next since some Windows 10-specific updates were made between the initial Windows runtime binaries and the latest release.

NFont’s creator Jonathan Dearborn has been running test apps I’ve sent him and sending back updates to try, and so far it seems we’re nearing a solution. Thanks for being so responsive, Jonny D!

The main major issue is signing my game’s binary. Windows 10’s SmartScreen puts up a warning about how they have protected your PC because they prevented the app from starting. It shows the binary as coming from Unknown Publisher.

That’s scary. I need to look into how to make it less scary. Does it require buying a code signing certificate, or is it similar to how Android’s code signing works? I don’t know yet, but I’m looking into it.

The other issue with Windows is that saving the game is sloooooow. In my game, I persist changes each time the player makes a major decision. Basically, if you click a button that switches to a different screen or causes something to happen in-game, I save so that if you shut the game down and reload it, it takes you back to where you were.

My Linux-based and Android-based builds are zippy. I can click, click, click, and any changes are instant. As a result, the game has been feeling very responsive despite the lack of a real-time need for it.

My Linux-based system does not have an SSD drive, and my wife’s Surface Pro does, and yet her system takes forever to save a file.

So on Windows, it feels less like click, click, click and more like click, wait, see screen update, then click. Because of the delay, sound effects are playing too early as well. It’s a lesser experience on Windows.

I haven’t ever needed to do multithreaded programming before as a single thread was usually plenty for the work I’ve ever done, but now I am wondering if I should spin of a thread specifically for writing to a file due to this issue that seems to be Windows-specific.

How Much Longer?

Ok, so there’s some technical issues, and some are easily surmountable, and some require some more investigation, and it’s possible there are some I haven’t run into yet.

Since Android seems the simplest to release, perhaps it goes into the Google Play store first, and I worry about the Linux and Windows versions later.

But I do not want this three month project to get to the ninth month before its first release.

The good news is that the next project will have a much clearer release plan, and many of these issues will be already solved. B-)

I started a three-month project at the beginning of the year, and I’m now in the eighth month. I reported on the reasons why it was taking so long last month.

But I’m feeling pretty good about it, and while I still have some balance issues to work out, and it’s a bit ugly, I’m preparing for the actual release.

The thing is, I haven’t really done a serious release before, and since I want to do a simultaneous cross-platform release, I’m finding issues unique to each platform.

The platforms I currently support:

• GNU/Linux
• Android
• Windows

What I want to support:

• Mac OS X
• iOS

I’ll start with Apple platforms, then talk about the environment I use natively. Other platforms will be discussed in the next post later this week.

Mac/iOS: no development or testing environments

I would love to create a Mac port. I know it is theoretically possible to create a cross-compiler to generate a Mac version, but it seems I need Mac-specific libraries, which requires owning a Mac.

I don’t own a Mac, and while I know of virtual Mac services you can subscribe to online, I haven’t bothered to look too seriously into them. I would also like to be able to test the game, and so I would need to use a Mac in order to see how it really runs, especially after running into the Windows-specific issues above.

As for iPhone or iPad, I’m in a similar position. I don’t own an iOS-based device. As I’m using libSDL2, I know it is possible to port to it, even without a Mac, but I would need to look into how to do so, and I would still need to invest in the devices to test on.

I am saving up for these things, but at the moment I don’t have them and I don’t want to spend time on them until I know what I’m doing.

And in the past it’s been difficult to hear back from people willing to be paid for porting a game for me, and volunteers have had difficulty figuring out how to put my project together on their system. I might look into it again, because that was years ago, and it’s a different world today.

GNU/Linux: distributing dependencies and architecture compatibilities

I develop and test the game on my Ubuntu GNU/Linux system, and the main thing to worry about there is that I can distribute the game and have it work out of the box on other distributions.

My game uses libSDL2 and related libraries. While I installed them on my system using my package manager, I can’t assume that my customers will have them installed as well.

Basically, I need to build custom dependencies, as per Troy Hepfner’s excellent article series on Linux Game Development, and then distribute them with my game.

Quite frankly, rather than worry about an installer to put everything in the correct locations on someone’s system, I think providing a basic tarball might be fine. Rather than provide .deb or .rpm or customer shell installers for each type of system, and then worrying about following the correct Linux Filesystem Hierarchy Standard, you allow the player to put the game in the directory of their choosing, extract it, and play.

But then I need to worry about how the tell the system to load the libraries. Running an application on Windows, the system generally looks in the local directory for libraries to depend upon. Unfortunately, Linux-based systems don’t do so, and while there is a way to point it towards your libraries using the LD_LIBRARY_PATH environment variable, I also know that it is frowned upon to do so due to the security and compatibility issues it can introduce.

On the other hand, many popular commercial games on my system do just that. For instance, looking at the directory for Don’t Starve, I see:

$ cat bin/dontstarve.sh
#!/bin/bash
export LD_LIBRARY_PATH=./lib64
./dontstarve

The fact that it is in this shell script wrapper is better than the original concern of changing the default environment variable in a more or less permanent way, which can cause version conflicts and such. It’s your program. You know what it needs, and any other applications that run will not be affected.

Still, supposedly the better way is to tell your binary at build time where to look, which isn’t very difficult. It requires -rpath=\$ORIGIN/[directory where you put your libs]. $ORIGIN expands into the directory that your binary is located.

So if the extracted tarball would have the following structure:
– foo-bin
– libs
– libfoo.so
– libbar.so

Then I would build foo-bin with -rpath=$ORIGIN/libs.

Of course, now foo-bin MUST be in the same directory as libs, but in practice, it’s fine. When was the last time you moved parts of a game’s files to different relative locations and expected it to continue to work?

I’m sure there’s issues with this approach as well, but with these two approaches, there’s plenty of precedent.

The only unknown I have is dealing with 32-bit vs 64-bit systems. Ubuntu has multiarch support, but I’ve seen comments on forums about people not being able to run an application due to architecture issues.

Don’t Starve distributes separate 64-bit and 32-bit builds. FTL, on the other hand, distributed both the 64-bit and 32-bit binaries and libraries together, and using a shell script, it determined which platform you were on at runtime to point LD_LIBRARY_PATH to the appropriate directory.

And other games distribute all desktop platforms together in one file, so if you bought the game, you bought it for Windows and Linux and Mac, whichever one you wish to play on. I like this option, especially since I hate the idea that I have to pay for a game twice in order to play on two different platforms.

I know some companies make their living by porting games and then selling them directly, but it’s not a business model I prefer.

Next time

In the next post, I will talk about issues specific to Android and Windows.