Permalink
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
476 lines (310 sloc) 25.7 KB

As you might or might not know to make Onikira: Demon Killer we use an engine called Duality and a few weeks ago we started to port the runtime to android.

Documenting the port of Duality game engine

Starting out

As we started we made a list of things that we needed to look at

  • Shaders
    • GLSL
      • All shaders need to be update to min #version 300
      • Fixed function methods like ftransform need to be removed and replaced with manual multiplication of transformation matrices
      • Shader in and out variables need to be defined for passing data from the app through the vertex shaders and into the fragment shaders
    • Engine
      • the engine should by default pass a bunch of predefined variables to each shader
      • predefined variables should be defined for common things like modelViewProj
      • the Shader class should try and pass all common variables to the contained shader at bind time
      • the engine needs to have a central place where common variables live that can be written to by the draw device when variables change (BeginRendering most likely) and read from by the shader classes.
  • Rendering
    • Currently uses fixed function vertex attribute bindings in the Vertex* classes. These need to be updated to use generic vertex attributes.
  • Windowing
    • need to specify the min GL version when creating the GameWindow class (DualityLauncher)

These were a conversation starter.

First steps

OpenTK

The first thing we did was to try to get OpenTk running. Here the awesome Dean Ellis gave us a huge hand by talking to us and then pushing some code to the repo.

Duality uses a fork of OpenTK, so to port it we forked too here.

  • add openTk.Android project
  • add a sample project to know everything is working
  • ...

Farseer

Duality depends on Farseer so we also had to fork this project. The process so far has been super simple

Difficulties

One of the main hurdles we found so far was to

Diary

Since we started I have been keeping a "diary" of what is going on for future reference. I want to try to document the process to hep any other porting efforts.

12th May

Today we finally got a texture rendering and using OpenGL 4.4 via OpenTK. We also decided that all the repositories will have their master untouched (as it was from any original fork) and that all android related changes will go into the branch 'android-port'

13th May

Today we are working on other default shaders rendering and input.

14th May

Today I am the only one working on the port... boo. I had to create a special project to use some hardware related jar and that brought me into the fun realm of consuming a jar from C#. To do that first I created an Android Java Binding Library following the steps here I was able to build this however it wasn't working as reported here

15th May - 31st May

... forgot to document until now (sorry :( ) we made progress on the library binding, in the end we wrote the binding code by hand :/ Can't remember what other changes happened. Re-booting the diary

1st June - input clean up

Cleaning up the input interface, the question here is should it go on the android branch or on the master? it seems like the android branch would be a better place, as it will.. ideally in the future be merged into the main

Maybe knowing how things depend on each other would help dependecygraph

2nd June - Input implementation

So after a refactor I noticed there were several not implemented methods, noticed a pronounced lag on the inputs, investigating... At the same time trying to figure

a) Does Xamarin android support [Fast Memeber] b) Will Binary serialization just work ( it seems like it should) c) how to make the controller/gamepad work via USB

3rd June - Answers to some questions

We were wondering

  • Would Binary serialization work? so I created a scene, serialized it to Binary and tried to run it on the android device, it worked :D
  • Does Reflection work on android? I tried this as well and I was able to reflect on a type and create it. Also spiked FastMember support and it works
  • Started Fmod integration

4th June

  • Working on the Fmod integration:
    • Created binfing library with no issues

    • Adding the *.so (wasn't sure what was the architecture of the target hardware) to find out you can run adb shell cat /proc/cpuinfo

    • If adb gives you errors about too many connected devices, make sure no emulators are running :D

    • in the end the jar file was not necesary, however I used one of the *.so, and you needed two :( the second one was on the fmoddesignerapi

    • for the moment the project is loading from a directory I copied to the device and it's hardcoded,, loading from assets doesn't work on the ways I have tried

5th June

So the fastest way to deal with the asset problem has been to save the file to the device on init/load, not loving this (suggestions welcome) but it does work.

Now I am working on making DualityScripts also be able to compile the C# and F# scripts also target android. Initially I was wondering if there was a way to target android with F# Compiler Services, but not the case, so I am just going to go ahead and create an android project that will be edited programatically to add all the script files and also edit the references and then build with msbuild :/

6th June

Today I spent some time fixing a bug on the scripting plugin project, it was important to fix it because I needed to make sure that we had a good way to build the scripts that live in Data/Scriptsthe tricky thing here is that we need to be able to build for multiple targets in different conditions. So today I fixed that bug and went over what was the current state of how we build al these and determined that I can use an fsx file to build this but will need to rely on the existance of an android project file per language to produce the assemblies, so it all points to refactor the current script so that we use FAKE so that there is clean control over the targets

7th June

As I was saying yesterday this build script needs to work on the following environments

  • teamcity build
    • target PC on release more
    • target Android on release mode
  • local build
    • target Android on any mode

The hard thing about building these is that these DualityScripts are resources (ie xml files) that contain code and have Duality related dependencies

I know this is not stricly part of the port but it is part of a set of problems that present itself during this process and I think it is important to make sure people working on the project can work as effectively as possible with as little disruption possible

  • created the empty C# and F# projects
  • disasembled current script into Fake targets this is a little harder than it might appear at first because of paths also I am hoping the build fails when something goes wrong :D

8th June

Checking paths and verifyign the build fails as it should (I also had to fix other unrelated issues )

9th of June

  • Add build script for openTK to Teamcity (builds and pack and push to myget)

Build info

  • for running xamamrin stuff on a build server via the command line you will need a business license
  • you need to run a tool called xamarin-componenet.exe login <your emall address> so that the server knows your license info (found this on the Greg Shackles talk about CI for xamarin)
  • nuget can be a PITA , but we all knew that

10th june

  • Package openTK into a nuget package
  • Create other builds some gotchas:
    • Component model .exe :(
    • When using FAKE and TC you ll need a TEAMCITY_VERSION env variable
    • For nuget packages targeting android use target="lib\MonoAndroid
    • don't mess up your paths :(

11th june

After some investigation, fixed the issues with package creation, just to discover issues with MyGet too :| my idiocy as I aparently managed to edit out a part of the url :( (uses package at the end when pushing the package) Creating farseer package now, also created Duality package

12th june

  • Created the Fast member build and package
  • Scrapped? copied? the teamcity server with a tool called Htttrack so that I can show current status in http://batbuild.github.io/PortStatus/ not happy with the results Need to come up with a plan for the rest of the port.

13th june

Starting attempt to compile game project (code name Honourbound and this is the name I ll use here to refer to it) Need to make the following available as android libs:

  • DisciplineOAk
  • Flow
  • spine-runtime

Maybe

  • Exceptionles
  • scriptingPlugin
  • protobuf-net
  • lz4
  • OgvVideo Player

Some notes tht I forgot to note before:

  • VS + resharper will show code as failing or missing references when in reality everything is fine, the compiler will tell you the truth
  • Sometimes something will not run on VS and will run and deploy fine on XS
  • Nuget packages do work on android projects however they need to have lib\MonoAndroid as their target.
  • On android, no resolution changing code
  • On android, no editor or debugging support code

15th June

By port I mean: created project and build scrips and build on TC and it builds and pushes a nuget package So I ported the following:

  • Spine runtime
  • Mercury particle engine (cross fingers it works :D)

20th June

Made the necesary classes available from Duality Scripting for Android and made the build availalbe

22nd June

Made changes to OgvVideo player so that it would be availabe from Android (ie compiles, but wont play video, for sure, bu tI gueess we ll get there)

23rd June

  • Checked asset size restriction is not a problem
  • added the msbuild target so that it will copy files to /obj/ (to build the apk) linking won't work as the files need to be there

To add data to the apk using a msbuild task that copies the data from DualityData folder (path dependency) to obj/ as the data needs to be present (linking is not enough)

#25 June

Made honourbound android sln (only contains coreplugin and gameplay core plugin) I wasnt sure if removing the tests was a good idea, maybe It will bite me in my arse but... only one way to find out right?

26th June

Patching up the ogv video player because I am pretty sure we dont need to use it in production.

Now trying to figure out how to load all assets and plugins and sounds and videos to a running android app.

Some "fun errors"

  • On the android manifest the vesion number needs to be an integer (there is also version name which seems to be the "real version number"
  • When debugging if you are inspecting something and the app crashes it is mostly the app if you see Mono.Debugger.Soft.VMDisconnectedException: it seems that that means that the error is on your side of the app
  • File.Exists always fails https://forums.xamarin.com/discussion/6286/file-exists-always-fails

27th and 28th June

  • realized some changes to duality hadn't been merged (from Andrew's branch) so that was merged
  • File cache created on start-up very slow , 5 seconds for 12 files :(
  • GC pauses for 1ms , this will probably be a problem
  • Fix mipmap problem that made rendering not work correctly
  • FileHelper abstraction

Rendering with mipmaps problem mipmap problem

rendering Imgur

29th June

A few days ago I had created a solution that will server as the place to collect all content and plugins . This solution would have an MSBUILD task that will collect content (Data folder in Duality) from a location and other similar things (sounds, videos, etc) from other suitable locations package it and deply the application to the console The problem with this is that the mdbuild task is very slow so this workflow might be acceptable for the build server tho not so much for the testing and running of the game. (even with the flag to avooid unchanged files)

There is a good explanation at Infinite Space blog about it. https://gist.github.com/tylerchesley/6198074 Fmod spike now copies the project from assets to somewhere in the drive (I am using SpecialFolders.Personal, is this a good idea?) and it is all working .. or at least I am playing sounds ...yay!

30th June

Today I

  • ...made it so the game project will have a post build step in the android project similar to the one on the pc project where the binaries are copied to a specific folder
  • ..streamlined the fmod spike and started to try to add it to the launcher project, however I hit a problem with dependecies :(
    • at the start I thought I didn't have the correct files, so I checked that ( there was only one asssembly that wasn't following the nameing convention and that was a little confusing, but now that is sorted)
    • checked that the assemblies where marked AndroidAssets and calling load for each of them results on a series of errors: FileNotFoundException :(

1st and 2nd July

The problem today and yesterday has been about loading stuff, at some point I had a problem with Fmod not loading the required *.so because some of the DllImportwhere not poninting to the correct place so code like this was needed in fmod and fmod_event

#if WIN64
        public const string dll    = "fmod_event64.dll";
#elif __ANDROID__
		public const string dll = "libfmodevent";
#else
		public const string dll = "fmod_event.dll";
#endif

the it was time to load resources, and we are copying both plugins and fmod stuff to the local disk, Dean did advice us against this, but couldn't find a workaround , anyway, when copying all there we stil need to do Assembly load to load th plugins. In Duality there is a good few places where Directory.EnumerateFiles with pattern is used, so I had to implenet something like this for our cache I am sure there is more to do regarding loading resources, however at the moment game plugins are loading and that is pretty cool.

3rd July

  • Adding missing default resources
  • removed some dependecies from nuget packages
  • Added font rendering :)

4th July

Fixed some bugs that helped getting music playing via FMod, this also meant that the main plugins (ie from the game) are working, or at least the types are loading and there are no errors when that is running. However when the song is playing the cats are not bounding

5th of july

So there is a problem. It seems to be the performance but as always , verification is needed. So there are a few otpions whern it comes to debug android

  • NVidia Tegra system profiler 2.2 : in par with other (pretty aewsome) NVidia tools tho it doesn't point to code (it ponts to the so)
  • Xamarin profiler: it is an ealy stage of this product and it shows, (crashed on me about 20 times ) and sometimes it will report 0 time

0 time then the other option is to use Duality profiler, but when I went to try that, we didn't have the default fonts...

  • Added default font, thou they seem to have some sort of dependency on a checkerboard texture
  • Adding default text

Video player with OGV notes (from Andrew)

  • found a repo here https://github.com/jhotovy/android-ffmpeg that was all ready to go. - Installed MinGW and MSYS. - Had to change Project\jni\settings.sh to point to NDK directory. - Added --toolchain=arm-linux-androideabi-4.9 --system=windows-x86_64 to the Project\jni\create_toolchain.sh script - ./configure and ./make for each of ogg, vorbis, and theora

Compile TheoraPlay against Android versions of libogg, libvorbis, and libtheora - found anroid-cmake here https://github.com/taka-no-me/android-cmake

6th July

  • Revise the default pixmap problem,
  • continue profiling to achieve stable framerate

7th July

  • finished the script that builds PC and Androidn builds of the assemblies for dualitySCripts (for both C# and F#) :D

8th July

  • there was actually a problem left on the script (ie show errors when the scripts should fail)
  • Created a MsBuild task to go over all the files and create a record on text
  • use the file as a cache
  • Andrew got the video integration using OGV working :D (running on the console)

9th July

  • use file as the cache (down to 16ms from 5 seconds)
  • performance

10th July

Fixed ogv player build (so that I can depend on the packages)

11th July

  • fixed bug on cache that prevented any file to be found as paths weren't normalized
  • everything is broken because reasons (...)

12th July

It gets to the point where I am not sure if things are broken or just too slow... tho too slow is kind of broken too ... Anyway, today I was trying to put togheter everything that worked, I found a little problem on the AndroidHelper class (it wasn't creating a directory if there wasn't one, but now it is) anyway.. it seemed like things were slow but working on then I thought it would be a good idea to integrate the input into Duality.

The input is something we worked with about a month ago, it was really painful to get working and we knew there was some issues with it, namely lag, significant (a few seconds) lag. Imagine my surprise when I go to run the spike and it does not take controller input..at all. Hours debugging to try to find out what the hell is going on :(

The input library lives in a *.jar specific to the manufacturer, and we patched togheter some jni so that the input with the controller worked. It is important to get this right because our game has rumble, and we got the controller rumbling before in our spike

At the same time, Andrew was working on fixing a bug on the IndexRenderer but got stuck for hours because he couldn't debug, after hours ... the solution was to uninstall Xamarin Studio and install the version I have running (he was on 5 somethihng I was in 4 something) as everything else was exactly the same on the two machines (including hardware) but I could debug and he couldn't.

23rd July

It has been really hard to write here, between preassure to get the build of the game ready for publishers and some unforseen problems (like the input just not working) ...

Anyway, in the days in between the last post and this one, it had become obvious that the scripts were not running, so as of today they do actually work, yay!! Other things that are done

  • Render targets run on android
  • scripts build to android somewhat reliably ( some of the problems of scripts not runnign came from there)

I copied FSharp.Core from where xamarin is using it locally, i really hope that is cool :D

The whole build seems to handle errors very poorly, tipical workflow would be that the failure reported is several layers disconnected from what the source of the problem is, for example, today I had a small samnple and tried running and this is what happened:

  • all assemblies (including dynamic assemblies ) loaded without errors, oninit method runs but screen is black
  • try to check is Scene.Update is running but I am not hitting a brekpoint
  • (head scartching... ) maybe the deubg dll is not there.. check and it is...
  • put a breakpoint in DualityApp.Update() (see code below) and the debugger would just get to OnBeforeUpdate() and exit
public static void Update()
{
	isUpdating = true;
	Profile.TimeUpdate.BeginMeasure();

	Time.FrameTick();
	Profile.FrameTick();
	VisualLog.UpdateLogEntries();
	OnBeforeUpdate();   //<-- HERE
	UpdateUserInput();
	Scene.Current.Update();
	//...
  • turn on exceptions to see what is going on... a bunch of the "normal" exceptions appear but nothing telling, so I debug OnBeforeUpdate and realise
private static void OnBeforeUpdate()
{
	foreach (CorePlugin plugin in plugins.Values) plugin.OnBeforeUpdate();
}
  • I realize our game, project code honourbound (the project is internally called honourbound, because it was the name of the game before some other game was released with that name :( ... ) was failing on a call to renderTargets,
  • so, I relize that the resources with the render targets weren't there , add them
  • still doesn't work, relize that the textures are also not there... add them
  • for some reason when I debug the project it hangs after loading all render targets ... I see no errors so I tr just deploying without debugging (because why not) and it works, the script actually runs, I am half thinking that maybe it doesn't work that maybe there is some sort of trick that makes it look as if it did work, but hey,

I will leave it here for today because it is less depressing than that some of the other days...

24th July

so today once again I try to launch the game, knowing full well it wont work but wondering what new and imaginative ways this will fail... So I created a new small sample that has a scene with a texureRenderer that after some time (via a script with some coroutines) transitions to another Scene.

But before that some quirks about working with Xamarin and Visual studio

  • if you are debugging on the device, everytime debugging on Xamarin Studio(from now on XS) will be faster than VS.
  • if you are trying to debug a refernced assembly on XS good luck that doesn't work
  • if you set some new keybindings, they tend to magically disapear. (i set some similar keybindings to the ones I used on VS and they key reverting to the original XS, this is very annoying bug)

So far I managed the hang not only VS but also XS, going though the errors .... I decided to unplug the console...

25th July

Definetly a much better day, I started of by repluging the console, this made me think maybe I should restart everything [acumulated state and whatnot] and the hangs were gone, I wish I was joking, my sample with one scene with a script on it that waited for a few seconds and then transitioned to another screen worked :| I mean, this is great but also very un-helpful

So this is an up to date list of things to do

  • faster development by compying the content to the device (otherwise it takes a few minutes to be able to start debugging.
  • Compression related issues. There are areas of the game like tiles, and other content that have been zipped for performance however when I started the port, I simply avoided those areas, knowing that we will have to think about this later
  • Input. As of today not working at all. Also integrating it into Duality is necessary. Perhaps a layer of abstraction would be wise here.

One important thing done today was to merge all latest changes to both Duality and the honourbound project , some of these changes should aid on performance.

Today trying the input again. For some reason I think about trying it in the other console... because it feels as if maybe we have inadvertidly changed something , some original setting... so we pair the controller to the other console and tada, the input sample that was working...works. I notice that the simbol when pairing is different to what I get when I attach the controller to the console , and that we used to get this symbol before and not anymore ... so I try in the original console to unpair the controller (that is usable to move around the menu and the sample controller applicaiton that came preinstalled in the console) and pair it again, and run the sample that we wrote to find it working again!!! The problem was that the device was not propoerly (what does this actually mean?) paired. Relif tho, at least now we can begin to understand how to deal with input. The lag is still there as a problem to be solved. and in having this problem we read a lot of the code in the controller library provided with the console.

At the moment Andrew is working on the compression related issues. Will add here when he is done

26th of July

I am thinking that attempting to integrate the input to Duality might make sense. And we did, to that effect we

Android zip support

  • Wrote a wrapper to make ZipOutputStream, ZipInputStream, and ZipFile collectively resemble the ZipArchive class from .net 4.5. The class is called AndroidZipArchive.
  • Documentation suggests that ZipFile is slow on first access because it builds an in-memory index of the contents of the zip. Since it's only used for random access to files, it isn't created when the AndroidZipArchive class is, but only when GetEntry is called for the first time
  • AndroidZipArchive.Entries uses ZipInputStream instead of ZipFile as that is supposedly faster for iterating through zip files.
  • To support the .Net 4.5 ZipEntry.Open method when in 'write' mode required writing a simple (and minimally implemented!) custom stream implementation called ZipOutputStreamWriter that simply forwards writes to an internal ZipOutputStream instance.
    • The game uses conditional compilation to pick the right IZipArchive implementation.

###27th July

Implemented the remaining keys for the controller , I am a little concerned because it doesn't have an up or down, it is either on or off Bring everything togheter

Summary

We are porting Duality runtime to Android, all the code is available on this user on the respective android-port branch of each repository. If you are interested in this and want to help or have questions, let us know.