Jenkins, Xamarin.iOS, & Calabash -- putting it all together

We've recently begun writing automated UI tests in Calabash for our main Xamarin.iOS project. While the tests are still under development, I decided to make a Jenkins job to fire off the the UI tests whenever a change was detected on the branch where the tests were being developed.

It took a little more than I was expecting to getting it running, so it is going to become the 2nd article in my impromptu series about Jenkins and Xamarin.

Jenkins in this case, is running on a Mac Mini running the most recent version of OS X (10.9.2) as a slave:


I like this configuration because it makes it easier for me to test -- if I can get it to run via SSH, then I can get it to run in Jenkins. In this case I think it may have also added an additional layer of complexity, but I learned a couple of things along the way.

My first step was to checkout the branch where the Calabash tests were begin written, and run all of the build step manually in the Terminal via the UI. A sanity check if you will. The commands looked something like this (most of these commands are explained in a previous article.

#Unlock the keychain
security unlock-keychain -p \"My supersecret keychain password\" \"${HOME}/Library/Keychains/login.keychain\"

#Set some environment variables to make it sort of like a Jenkins env
APP_VERSION=$(date +%Y.$(((($(date +%-m)-1)/3)+1)).%m.${BUILD_NUMBER})

#Run through the Info.plist modifications
/usr/libexec/PlistBuddy -c \"Set CFBundleVersion ${APP_VERSION}\" Info.plist
/usr/libexec/PlistBuddy -c \"Set CFBundleIdentifier com.ourawesomeapp.${CLIENT}\" Info.plist
/usr/libexec/PlistBuddy -c \"Set CFBundleURLTypes:0:CFBundleURLName com.ourawesomeapp.${CLIENT}-handler\" Info.plist
/usr/libexec/PlistBuddy -c \"Set CFBundleURLTypes:0:CFBundleURLSchemes:0 ourawesomeapp-${CLIENT}\" Info.plist

#Clean the solution
/Applications/Xamarin\\ build -c:\"Debug|iPhoneSimulator\" -t:Clean OurAwesomeApp.sln

#Build the solution
/Applications/Xamarin\\ build -t:Build -c:\"Debug|iPhoneSimulator\" OurAwesomeApp.sln

The next step was to issue the cucumber command, which ultimately failed (after spending some time laughing at me). Turns out that building for the simulator doesn't cause the app to get pushed to the simulator, which is what is needed for Cucumber to execute correctly.

At this point, I opened the solution up in Xamarin Studio and built again, this time with the deployment to the simulator. I kept my eye on the Build Output screen, which ends up outputting most of the commands that are issued behind the scenes when you tell Xamarin Studio to build. I found a command that seemed like it would do the trick, so I copied it out into an editor, and tweaked it to make it suit my needs.

/Developer/MonoTouch/usr/bin/mtouch -sdkroot \"/Applications/\" \
--cache \"${WORKSPACE}/obj/iPhoneSimulator/Debug/mtouch-cache\" --nomanifest --nosign \
-sim \"${WORKSPACE}/bin/iPhoneSimulator/Debug/\" \
-r \"${WORKSPACE}/MTSplitViewLib/MTSplitViewLib/bin/Debug/MTSplitView.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/System.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/System.Xml.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/System.Core.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/System.Web.Services.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/System.ServiceModel.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/System.Runtime.Serialization.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/Mono.Data.Sqlite.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/System.Data.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/monotouch.dll\" \
-r \"/Developer/MonoTouch/usr/lib/mono/2.1/MonoTouch.Dialog-1.dll\" \
-r \"${WORKSPACE}/Assemblies/FlyoutNavigation.dll\" \
-r \"${WORKSPACE}/Components/calabash-1.5/lib/ios/Calabash.dll\" \
-debug -nolink -sdk \"6.1\" -targetver \"6.0\" \
--abi=i386 \"${WORKSPACE}/bin/iPhoneSimulator/Debug/OurAwesomeApp.exe\"

That's a whopper of a single command, but as I understand it from the mtouch documentation, it pretty much is just defining the build environment -- telling mtouch explicitly where to find the SDK to build against, which assemblies to include in the build, etc. The command that I got out of the IDE had a lot of specific paths, and I did a find/replace job on them so that it could all be relative to the Jenkins ${WORKSPACE}.

In the future I'm going to revist this command. It makes the build job brittle in that if we remove, say, 'MTSplitView.dll' from the main solution, then this command will break the build process. When I find a way to make it more flexible, I'll update this post.)

After running that command in the Terminal, and then starting the simulator manually, I did find the newest version of the app in my simulator. Yippee!! Now cucumber would run and all of the tests would run on the build server.

Seemed as if I was in pretty good shape, so I ssh'd into the same build server, as the same user that I was logged in as, and ran the whole thing again. Everything ran perfectly up to the point where cucumber started. Then it started laughing at me again.

After some Googling & testing, I figured out that the very first time I would walk through my process in the UI, I would get a prompt to enter my password. It always happened after I would execute the cucumber command, and after the simulator had started, but before the tests started. I didn't pay much attention to it the first time around, and since I resolved most of the problems in the UI in a single session, I never got prompted again.

My SSH session was trying to prompt, but had no UI to display the prompt on, so it was just choking, waiting for me to authorize cucumber's request to control the simulator. There are quite a few threads about this on the internets, but they were all a little dated, and none of them worked quite right for me (which is what inspired this article).

The first step was to make my builder account a developer:

/usr/sbin/DevToolsSecurity -enable

That command is in my history for that day a couple of times. Sometimes without the fully qualified path, and sometimes with sudo. I don't have great notes so I don't know exactly which one did the trick, but this output let's me know that one of them did it:

~$ /usr/sbin/DevToolsSecurity
Developer mode is already enabled.

The next step, according to what I had managed to scrape up from Google, was to allow the taskport privilege to be assigned without prompting.

Before executing this next command, I did some research on it. I encourage you to do the same. As I understand it, the following command will allow a process running as your user to take control of another process running as your user. While it is exactly what we want in this case, it is generally something that is allocated on the fly. This command will make the temporary allocation permamnent. There is a chance that this could be used against you. Consider yourself warned.

sudo security authorizationdb write system.privilege.taskport allow

The output from that command was encouraging:

YES (0)

so I ran cucmber again (from SSH), and lo and behold, it worked. Needless to say, I was very excited. I logged out, and ssh'd in again, and it worked again. Which lead me to the conclusion that Developer mode only needed to be enabled once, and the security authorization command did not need to be reissued everytime.

I added this command as my last Execute Shell in the build process in Jenkins:


and fired off the job. It ran without issue.

Next someone asked if there was a way to get a report from cucumber, and it turns out that there is. I modified the last step to

cucumber -f html -o CucumberResults_${BUILD_ID}_${BUILD_NUMBER}.html

and then used a post-build Send build artifacts over SSH to copy that html file and all of the screen shots over to an apache server that indexes the reports:


Pretty cool right? Clicking on one of those folders takes you right into the report for that specific build.

This introduced the need to do some additional cleanup at the beginning of the build, so I added a new Execute Shell as the first step of the build:

rm -f CucumberResults*.html screenshot_*.png

It's not very elegant, but it does get the job done.

I was ready for my final test, before I put this one to bed for a while. So I fired the job, and it failed. Ugh.

Turns out that one of the Calabash tests had failed, which caused cucumber to exit with a non-zero status. This cause Jenkins to consider the build failed, which meant that the report was not copied over, so we couldn't really see what had happened.

I decided to take the easy way out. I modified the cucumber step in Jenkins to look like this:

cucumber -f html -o CucumberResults_${BUILD_ID}_${BUILD_NUMBER}.html || true

which will cause the command to always return 0, thereby tricking Jenkins into always thinking that cucumber exited cleanly. There is probably a more elegant way to solve that problem, but for the time being, it works.

Now as the developers are working on bugs and new features, they are able to write the UI tests for Calabash, and submit them. For the time being, it is still all living in a separate git branch. Jenkins is monitoring that branch, and when changes are pushed to it, Jenkins fires off. Once it is all merged back into master, then Jenkins will be modified -- and we will probably begin running our UI tests nightly.

Of course it isn't quite as good as I want it to be yet. I still have the following things that I need to figure out (so stay tuned, it will be showing up here as soon as I figure it all out):

  • specifying the OS version to run in the simulator. Right now, the simulator starts with whatever the last version was run. I'd like to be able to have it run on every version of the simulator that is installed & that we are currently supporting (right now that is 3 versions of iOS).
  • making it run on an actual device. The first rule of app development is don't do all of your testing in the simulator. It's a good rule, and I want to follow it.
  • making it all work in Android.

In fact I have a lot of Android work yet to do...

Comments powered by Disqus