Friday, March 17, 2017

Detecting memory leaks on Android apps

More often than it should you find yourself wondering if your application is leaking memory and if so what, where and how.
A lot of questions that usually don't have enough answers.
There are many well know ways your application can leak memory. To name a few

  • static references to Contexts, Views or Activities
  • inner classes holding a reference to the outer class
  • anonymous classes, commonly used in listeners
  • drawables holding a reference to the Context
  • and many more cases

If you are interested in more detailed descriptions of these common ways your app can leak memory this article will help you.

Also, as there are pitfalls there are tools to help you find them. Again, to name a few

  • Android Studio monitors
  • dalvik and art GC log messages
  • heap dumps
  • dumpsys
  • etc.

To do an effective use of these tools and uncover memory leaks you should stress your app, running it for a while, or forcing continuous invocations of Activities, perhaps navigation back to the Home screen and launching the Activity again and again. While you can do this manually and the most obvious cases will be detected, it is always more efficient to automate these steps so they can be run before and after to verify that you actually solved the problem.
And what better than AndroidViewClient/culebra to automate these steps as we have discussed so many times in previous articles. Precisely, one of the latest additions is the ability to capture dumpsys information and create plots with it.

This is an example of a method we can define to exercise our Activities including some extra steps like stopping the application to start clean and forcing garbage collection along the way (see complete source code here)

def __plot_dumpsys_meminfo(self, pkg, activity, method=None):"am force-stop %s" % pkg)
    for n in range(20):
        if n % 5 == 0:
                "run-as %s pgrep -L 10 %s" % (pkg, pkg))
        self.device.startActivity("%s/%s" % (pkg, activity))
        if method:
        self.plot.append(Dumpsys(self.device, Dumpsys.MEMINFO, pkg))'BACK')

With every iteration, we are collecting the dumpsys meminfo for the process, exiting the app by sending BACK and returning to Home, and at the end, we plot the chart.

We can easily see that while we are forcing GC the amount of memory used, the number of Activities and Views are constantly increasing.

As always, you can find more information about AndroidViewClient/culebra in its wiki at, about CulebraTester at and and if you have any question you can ask in Stackoverflow using

Hope this help you spot some leaks in your app.

Thursday, October 27, 2016

CulebraTester Calculator Demo

Android testing can be complicated, time-consuming, and tedious. What if it didn’t have to be?

CulebraTester provides a real-time point and click test recording through a web browser.

We are currently in private beta, so if you are interested in participating of this programme please fill the opt-in form and we will be adding your account as soon as possible. 

Monday, August 01, 2016

Turns your Android smartphone into a complete laptop

I do not usually mention third party products in this blog unless there is a very good reason. The Superbook certainly is one of those.  It is a smart laptop shell that provides a large screen, keyboard and multi-touch trackpad, 8+ hours of battery, and phone charging capabilities. When plugged into your Android smartphone, it launches our app to deliver the full laptop experience.

The idea of using the same core device for your mobile and desktop experiences has been tried before, but this time, the Superbook gives it a twist and uses a full Android desktop. This is mainly possible due to the huge improvement lastest Android versions received in areas as multi-window support,  Android for work and Keyboard Shortcuts screen that surely move  the platform in the desktop direction.

That was not true some years ago. 6 years to be more precise, when one of the projects I was part of implemented the same concept but in the opposite direction I would say.

A mobile computing device with a mobile operating system and desktop operating system running concurrently and independently on a shared kernel without virtualization. The mobile operating system provides a user experience for the mobile computing device that suits the mobile environment. The desktop operating system provides a full desktop user experience when the mobile computing device is docked to a secondary terminal environment. The mobile computing device was a smartphone running the Android mobile OS and a full desktop Linux distribution (Ubuntu) on a modified Android kernel.

The concept included also a Smart book where you could dock your device.

Unfortunately, this project was later discontinued.

As I mentioned, a lot has changed in these last 6 years on Android and devices are much more powerful now to support this approach so take a look at Superbook as it might change your mobile and desktop experiences.

Friday, May 20, 2016

No Espresso Test Recorder in Android Studio 2.2

After all the frustration you may have experienced after discovering that Espresso Test Recorder is not in Android Studio 2.2 as announced in Google IO/16 presentation (see you can have something to play with today. Sign up for CulebraTester private beta see opt-in form)

Saturday, May 14, 2016

CulebraTester Private Beta Opt-in

Android testing can be complicated, time-consuming, and tedious. 
What if it didn’t have to be?

CulebraTester provides a real-time point and click test recording through a web browser. This browser is connected to the Android device under test. Not sure what we mean?
Visit for details.

If you are interested in being part of the Private Beta please fill out this form.


Tuesday, December 22, 2015

AndroidViewClient/culebra vs. MonkeyRunner

More than 2 years ago I took a crucial decision in AndroidViewClient/culebra development plan and that was to free it from `monkeyrunner`, Jython and Chimpchat.

AndroidViewClient/culebra was liberated and starting with version 4.0.0 it does not require any other runtime environment than python 2.x (read announcement). It can be installed and upgraded using the corresponding platform tools like easy_install or pip and can be easily integrated into IDEs like Eclipse PyDev or Pycharm. It also improves speed, solves chimpchat bugs, and even provides a GUI whre you can automatically create tests or scripts without writing a single line of code.

Nonetheless, from time to time, I receive some questions or reports about problems with scripts created with `culebra` that are attempted to run with `monkeyrunner` or some other combinations. I take the blame for it. I failed at communicating that AndroidViewClient/culebra is a complete replacement and should not be used together.

In order to improve the situation I gave a very detailed, easy to follow, step-by-step answer to
Error of Script with MonkeyRunner and AndroidViewClient (Touch) on Stackoverflow, showing how you can create a test case that automatically starts.and Activity (Duolingo) , checks if some Views are on the screen, touches them and finally take the screenshot. All from the GUI.

I hope you find this explanation useful.

Monday, August 24, 2015

culebra: concertina mode

One of the latest additions to culebra is the concertina mode. Named after the concertina movement occurring in snakes and other legless organisms that consists of gripping or anchoring with portions of the body while pulling/pushing other sections in the direction of movement, this mode allows culebra to move across the UI.

You may have used android's monkey before. monkey is a UI/Application exerciser that generates pseudo-random streams of user events such as clicks, touches, or gestures, as well as a number of system-level events. It is commonly used to stress-test application. While it may serve this purpose, it's overly complicated to be used in other scenarios.

Then, it comes culebra concertina mode that instead of sending pseudo-random events, analyzes the content of the screen and randomly selects a suitable event or action for the also randomly selected target, normally a View.

For example, if the randomly selected view is an EditText, culebra enters some random text. However if the EditText turns to be a password, it sends random passwords, that is instead of just entering a normal sentence it selected sample passwords from a list like the infamous "querty".
Furthermore, if it detects the EditText is expecting an email address, by inspecting the id, the hint or anything else that may suggest that, it sends random email addresses like Of course, other Views, receive other treatment, like Buttons being clicked, Scrollables being scrolled, etc.

Additionally, if the content description suggest the application may be awaiting for you to talk, like 'Tap to Speak' or 'Voice Search', culebra will actually speak random text to the app (this feature is only available on Linux and OSX).

You should check this aspect, it's sometimes hilarious.

Taking advantage of the culebra GUI already described in previous posts, while it's running the randomly selected Views are highlighted. At the same time, the python script is being generated. You can save it, modify it and run the same tests again and again. It's worth to mention that commonly the script eliminates the random selection of components and thus it runs deterministically.

Run culebra in concertina mode

$ culebra --gui --concertina --scale=0.5

and you can obtain something like this, where the nickname of a new contact was randomly selected from the available Views and the random text "chaos reigned" was entered.

BTW, this screenshot was taken via dump as described in android: obtaining beautiful screenshots automatically.