Haxe 3, OpenFL, Arch, Android

haxeTom and I went through and installed Haxe 3 and OpenFL on our machines today.  OpenFL is the updated Haxe 3 version of NME.  These are a few notes, updating from my Haxe 2 comments, some of which are still applicable. All in all, the process seems smoother than it was for Haxe 2/NME on Arch… I’m excited!


First install Neko out of the AUR:

sudo packer -S neko

Then Haxe 3:

sudo packer -S haxe

Note that you can do those two steps at once. If you receive an error about an incorrect Neko package version, it’s because the Haxe package doesn’t denote the version dependency. Running them separately as above will force Neko to be updated first, and then Haxe.

At this point you should setup haxelib:

haxelib setup

Tom sets the haxelib location to be in his home directory, which is most proper. I leave it in the default /usr/lib/haxe, for no good reason. In that case you need to modify the install directory to be owned by your user:

cd /usr/lib/haxe
sudo chown -R joe .


Next, install OpenFL following their instructions:

haxelib install openfl && haxelib run openfl setup

When it asks for sudo permission, it’s installing a simple script aliasing haxelib run openfl to just openfl, so you can skip it if you wish.


export HAXE_STD_PATH=/opt/haxe/std

After that you should be able to create and run a demo:

openfl create DisplayingABitmap
cd DisplayingABitmap
openfl test linux
openfl test flash



Next, try to set up OpenFL’s Android support:

openfl setup android

That script will pull down the Android NDK, SDK, Apache Ant, and Java, but if you already have all that installed simply say no to each download request and enter the path manually.  You can then try to build a sample:

openfl create DisplayingABitmap
cd DisplayingABitmap
openfl test android

If you’re using a recent 64 bit android NDK, that will fail with a root cause of not finding g++:

sh: arm-linux-androideabi-g++: command not found

To begin with, in your haxelib directory, edit build-tool/BuildTool.hx. Specifically, the Linux parameter by line 1355 in the Android block should be altered from just the value linux-x86 to the following:

m64 ? "linux-x86_64" : "linux-x86"

The block will then look like:

      else if (defines.exists("android"))
         if (!defines.exists("ANDROID_HOST"))
            if ( (new EReg("mac","i")).match(os) )
            else if ( (new EReg("window","i")).match(os) )
            else if ( (new EReg("linux","i")).match(os) )
                          m64 ? "linux-x86_64" : "linux-x86");
               throw "Unknown android host:" + os;

That last ANDROID_HOST setting is what matters; as shipped, it does not point to the 64 bit toolchain correctly.

Then in build-tool/android-toolchain.xml, edit the toolchain version to 4.6. Edit line 7 setting TOOLCHAIN_VERSION as follows:

<set name="TOOLCHAIN_VERSION" value="4.6" unless="TOOLCHAIN_VERSION" />

In build-tool/ then run:

haxe Compile.hxml

OpenFL, specifically openfl-tools, defaults to Android API 8. If you have that installed, you should then be able to build the OpenFL samples for Android. You will though need to stipulate 64 bit hxcpp:

openfl test -DHXCPP_M64 android

To change the target Android API, in your haxelib directory edit openfl-native/1,0,6/templates/android/template/build.xml and change the target property on line 7, e.g., from:

  <property name="target" value="android-::ANDROID_TARGET_SDK_VERSION::"/>


  <property name="target" value="android-16"/>

OpenFL should then build, push, and run the target on your phone using this API version, stipulating the 64 bit toolchain as above. Hooray!

Success on my S3!

Success on my S3!

Gold Leader @ 50,000

goldleader-iconSometime today Gold Leader crossed 50,000 unique viewers.  Together they’ve played just under 76,000 times over the two weeks it’s been released.

Of course, “unique” is not a strictly literal term.  I’m not sure how Google Analytics determines uniqueness exactly, but people coming back after some time are clearly included.  In some sense that doesn’t matter much in terms of counting eyeballs for ads and such and there’s no other real mechanism without any sort of accounts or such, so that’s the kind of number that generally gets bandied about.

As noted by the stats on the main blog, the game remains much too hard: Only about 14% of all players are beating the first mission objective.  I can only imagine what that’s doing to the repeat plays, or presumed lack thereof.  About 4000 people have loaded the game and not started a mission. I guess it just wasn’t what they’re looking for.

Almost all of the plays actually come from the last week.  The first week after launch (Jan 16) the game was getting a steady thousand views/day or so on GameShed.  Once it went viral to all the ridiculously many Flash sites around the net though, the plays spiked quite dramatically (about Jan 22).  Views are trending down now as the game falls off their front pages, but I really have no idea what the lifetime is.  Hopefully it settles on some sort of steady number of plays and doesn’t bottom out completely though it’s hard to believe that won’t happen given the steady deluge of games hitting the Web.


Plot of game load events as captured in Google Analytics.

I’ve had some trouble coming up with numbers for typical Flash game plays so I’m not sure how this compares.  Many sponsors don’t allow developers to collect stats, and many developers are cagey about revealing what they do know.  Clearly it’s no Angry Birds or even a moderate hit, but it also certainly doesn’t seem to be the bottom of the barrel either.

Either way, that’s a pretty satisfying number and was actually my admittedly under-informed target: 50,000 players in two weeks.  50,000 people is a lot of people!  I guess it’s not “Internet scale,” but even accounting for some reasonable fraction being repeat players that’s still an absurd amount of people to have played my little shoot ’em up.

To put it in one perspective of interest to me, I’ve published research articles in several high profile venues, most notably in terms of numbers some of the major IEEE magazines such as Internet Computing.  An optimistic readership number I’ve heard for those higher profile venues is 5000 people, and presumably some fraction thereof reading any given individual article.  This is an order of magnitude more!  And it’s god knows how far from putting games up on the Web ten to fifteen years ago, when I was probably optimistically getting a very low couple hundred players, mostly through the Allegro community.

For the most part this is actually not as emotional as some of the early super enthusiastic reviews on FGL, but it’s still really cool and deeply rewarding.  It’s also somewhat educational.  Though fairly obvious beforehand, the data confirms that even for a lower profile game like this, the resources consumed would be pretty substantial for a single host.  If I moved the leaderboard or analytics off Mochi and Google, or provided any kind of multiplayer, online updates, or downloadable content, I would need some serious infrastructure to manage it.  Fortunately, I’m on top of it.  To the future!

Installing Haxe/NME on Arch Linux x86_64

haxeSome updated notes on Haxe 3 and OpenFL are now available.

I love Haxe and NME.  But they’re at times not the most pleasantly packaged of software.  There are some definite lapses in checking the released versions for packaging problems, the nightly snapshots often have problems, and few of the setup scripts will work if you’re not using an apt-based system (i.e., Ubuntu). Having just installed on two 64bit machines running Arch Linux, the following are some notes on getting it all to work.

Flash Player

You’ll have to install the 32bit multilib stuff in order to install the Flashplayer standalone debug player.  You definitely want this as the debug player seems to be the only way to get reports on null pointers and such in your Flash programs.

Enable the multilib repository by editing /etc/pacman.conf to uncomment:

SigLevel = PackageRequired
Include = /etc/pacman.d/mirrorlist

Then udate the repo DB and install the multilibs:

sudo pacman -Syu
sudo pacman -S gcc-multilib

Accept everything pacman asks in running that command. Then install the player.

sudo packer -S flashplayer-standalone-debug

Haxe/NME Toolchain

The Haxe/NME stack includes a number of components. You could try installing these from the NME install script, but it’s based on Ubuntu and several components won’t work even after you adjust for that. Instead you need to cherry pick some parts and replace others.  The rest of this sort of follows the script, but injects some variations and other steps.

The first thing to install is Neko. You want to do this from the AUR package, e.g.:

sudo packer -S neko

Installing from the tarball like the NME script does will result in a broken dependency with libpcre3. Manually compiling from source to fix that seems to be a rathole, and using the AUR package seems cleaner than jury-rigging a .so fix.

Next is Haxe itself. Tragically, you seemingly don’t want to pull this from the AUR package. I haven’t looked at it closely to confirm, but believe it pulls from SVN, which often leads to problems. Today on installing from that I get errors along the lines of “Missing Standard library.” (not an exact quote). Instead, download the stable release and patch up an install:

wget -c http://haxe.org/file/haxe-2.10-linux.tar.gz
tar xzvf haxe-2.10-linux.tar.gz
sudo mkdir -p /usr/lib/haxe/lib
sudo chmod -R 777 /usr/lib/haxe/lib

sudo cp -r haxe-2.10-linux/* /usr/lib/haxe
sudo ln -s /usr/lib/haxe/haxe /usr/bin/haxe
sudo ln -s /usr/lib/haxe/haxelib /usr/bin/haxelib
sudo ln -s /usr/lib/haxe/haxedoc /usr/bin/haxedoc

haxe /usr/lib/haxe/std/tools/haxelib/haxelib.hxml
sudo cp haxelib /usr/lib/haxe/haxelib

haxe /usr/lib/haxe/std/tools/haxedoc/haxedoc.hxml
sudo cp haxedoc /usr/lib/haxe/haxedoc

sudo haxelib setup /usr/lib/haxe/lib

wget -c http://www.haxenme.org/builds/hxcpp-2.10.3.zip
sudo haxelib test hxcpp-2.10.3.zip

Finally, install NME and friends.  Begin with the basic download.

sudo haxelib install nme

From here I have tremendous problems.  Nothing will actually work yet, including the rest of the setup process; I simply get errors from tls.ndll about missing libopenssl.so.0.9.8. Seemingly there is a persistent problem in the packaging of NME such that it’s not properly searching the appropriate 64/32 bit Neko libraries. At this point you’re standing on the edge of a huge rathole. Compiling NME from source seems to be fairly complex to setup outside of the assumed Ubuntu environment. Installing that version of OpenSSL doesn’t work because it’s 32 bit. After repeatedly screwing around with this at length on several major releases of NME, short of setting up NME from source I don’t think you have any option here but the ugliest:

cd /usr/lib
sudo ln -s libssl.so.1.0.0 libssl.so.0.9.8
sudo ln -s libcrypto.so.1.0.0 libcrypto.so.0.9.8
cd -

Now you can continue installing and hope you haven’t just set some lurking booby trap to slam you on some other project months down the road…

sudo haxelib run nme setup
sudo haxelib install actuate
sudo haxelib install svg
sudo haxelib install swf

At this point things should work. However, I had trouble getting some of the RocketHaxe demos to compile with NME 3.5.4. It produced inscrutable and inexplicable errors about exceptions creating the output files. Updating to 3.5.5 seemed to fix these; I have no idea what else it broke in return. You can upgrade to a nightly build by downloading a new version from the NME builds repository. Then have haxelib switch to that version, e.g.:

sudo haxelib test nme-3.5.5-165-g354e1e6.zip


Now you should be able to test everything out. Go into a random temporary directory and build a sample program:

nme create Addingtext
cd AddingText
nme build project.nmml flash
flashplayerdebugger Export/flash/bin/AddingText.swf

If that works, it works. If it doesn’t… Slaugther a chicken and call a priest or something. Please leave any comments you have about improvements to this process or other notes. Thanks!

Further notes about setting up Haxe for Android development are here.