Android Development Gotchas

Posted on by
2

Android Development Gotchas

This isn’t designed to be a definitive list of unexpected occurrences while developing for Android, but it is a list that I scratched down while I was picking up some skill for myself. It won’t save you from every issue but there may be cases where you can stop and think “Yeah, Dave told me that would happen”.

My development environment included Eclipse 3.5 with the Google ADT plugin and all testing was against a rooted HTC Desire and unroot Samsung Galaxy SII both running Android 2.2, development code targeted Android 2.1. Problems may be an issue with ADT, HTC, Samsung, Android 2.1 or 2.2, rooting or the specific hand sets 🙂

Java 1.6 on a Dell XPS 1530 running Ubuntu 10.10, if that matters.

Virtual Devices

I found these too slow and unusable.

Early on in development I gave up trying to use the the ADT virtual devices and tested directly against the target hardware devices. Luckily for me the application was for internal use only and would be released to specific clients with specific handsets, but I was disappointed at how slow the VMs were start and how slow they ran. The develop-deploy-test cycle was just too slow using the ADT.

Virtual Devices Cannot Support Bluetooth

ADT Virtual Devices cannot support Bluetooth, and as our application includes Bluetooth communication this was another reason ADT development was excluded.

Inconsistent Bluetooth Support Between Vendors

This was a strange one to us, and not one that we have (to this point) been able to resolve although my guess is that it is a problem with the HTC code.

Our application involved having a file sent to the phone using OBEX over Bluetooth. No problems so far. Pair the devices, enter the code, initiate the file transfer.

The Samsung Galaxy SII displayed the preferred behaviour, where the incoming file prompt the phone user to accept the file, but also provides a checkbox to always accept files from this device. If you don’t select the checkbox then you are prompted each time a file arrives, but you are still able to check the box later on.

When the first file arrives, HTC Desire asks if you want to allow the remote device to synchronize contacts, and whether you always want to allow this. If you don’t allow this, you aren’t able to get to the next step where you accept the file. If you don’t accept contact swap all the time you’ll be prompted each time. Regardless of whether you swap contacts or not, you must accept each file as it comes in. Accepting every file manually is not our preferred behaviour.

Android Apps Run on a Single Thread

Some of this cuts into Android’s Handlers, task scheduling and communication across threads, but it may come as a surprise during development that all standard processing occurs on a single Thread. This is something that you’ll want to play with and become familiar with as it may be surprising that UI performance is impacted by Runnables and Services.

I do recommend you have a play.

Creating new Threads is one option, although Android’s AsyncTask is the preferred mechanism for truly asynchronous processing that doesn’t freeze the GUI.

… and that Thread is a UI Thread

Following on from the previous point, the main thread is a UI thread and that that thread is only active when the UI is active. Sort of. Check the Android fundamentals processes and threads page, but just be aware that if you want something to process while (for example) the screen is locked, you need to be aware of the processing thread.

Beware Empty Elements in XML Layouts

If you’re like me, you like tidy XML and prefer an empty element (where applicable) to an element which is not ’empty’ but has no child elements.

	<!-- this element has no children -->
	<ListView android:layout_width="fill_parent" android:layout_height="wrap_content" android:id="@android:id/list"></ListView>
	<!-- this is an empty element -->
	<TextView android:id="@android:id/empty" style="@style/EmptyContent" android:text="@string/emptyResponseEntries" />

In this example, the TextView is empty and causes no problems, but the ListView will. If you make the ListView an empty XML element, you can write items to the List, but they won’t show up on the screen. There won’t be any error reported, it just won’t show up.

Can’t Read Property Defaults

Property defaults were also something I found perplexing. If you specify a properties.xml file, each property is able to define a default value. Likewise when reading properties via the SharedPreferences you can specify a default value if the property doesn’t exist, but the two default values aren’t related. You can’t read the default value as defined in the preferences.xml file for use elsewhere, the API doesn’t provide access.

Note that recommended properties management is via the PreferenceFragment

Notifications Can Be Empty

I found this one amusing. It is possible to create an empty Notification using new Notification(), specify a notification sound, and annoy users without providing any indication of where the annoying sounds are coming from.

installing postgres and tomcat on the mac

Posted on by
3

Back when I was setting up my Mac, I installed Eclipse 3.7.  I also installed Postgres and Tomcat so I could run locally.  Postgres was either unnecessarily frustrating or it required knowledge I didn’t know I was missing.  The whole shared memory thing should have been a clue.  The good news is that I think I did almost everything wrong one could so I have a nice list of problems.  I wrote this up two months ago, but waited to post until I understood better what was going on.

I learned I didn’t know where the applications folder is on the hard drive.   First I learned you can view the path of a file in the finder. Which is all well and good.  It tells me the path is Macintosh HD > Applications > Eclipse.  Awesome.  On a UNIX file system, a directory has a slash up front.  Ok.  So I’m now at a UNIX prompt looking for it.  No Mac HD at the root level.  Volumes looks promising so I cd there and see Macintosh HD as a choice.  Great.  And there’s even an Applications folder under it with my apps!  I found it without having to use “find.”  That wasn’t bad, but it was more surprising than I expected.  I also noticed there is /Applications which will be more convenient next time.

Also, the “one click installer that doesn’t tell you what it is doing” installs some things by unzipping to the directory you are in, some files to /Library and presumably some files to who knows where.  Now I see the scripts I was expecting.

initdb doesn’t have enough privileges as me so I tried root and got initdb cannot be run as root.  Ok so it’s a real UNIX install where I need a user to run as.  I wasn’t expecting this because all my previous installs on the Mac have shielded me from a real UNIX install.

I thought I needed to create a  postgres user to run the database.  In Apple’s system preferences it tells me an id with that name already exists.  But /etc/passwd says nothing about it.  Running

dscl . -list /Users UniqueID

does turn up a postgres id. I need to figure out how the Mac Directory Service thing works.  Anyway, now that I know the postgres users exists I did “sudo passwd postgres” to reset the password to something I know.  Turns out the “sorry” message when trying to sudo to postgres wasn’t from an incorrect password.  Eureka!  I needed to type “sudo su postgres” and then type *my* admin password to switch.  Which kind of makes sense because my regular id isn’t allowed to switch user but my sudo admin user is.

I then changed owner of /Library/PostgreSQL recursively to postgres user.  A nice easy step.  (sudo chown -R postgres /Library/Post*)

Finally, I could run initdb and get prompted for the new superuser password!

I then tried to start the database and got a complaint the lock file already exists.

FATAL:  lock file “/tmp/.s.PGSQL.5432.lock” already exists

HINT:  Is another postmaster (PID 94) using socket file “/tmp/.s.PGSQL.5432”?bash-3.2$

Tried rebooting.  Error still there.  Turns out postgres automatically starts up.  I went on to create a user and database using the usual commands.

Then I opened pgadmin to see if I could access the database.

Connected – maintenance db is database name.

Almost there.  I used the wrong password for the id I created in postgres.  (This matters because the coderanch integration tests assume a certain password.  And because the one I picked I will never remember.)

alter user postgres with password ‘better_password’;

Importing data from file went smoothly.

Ongoing Little problems

  1. Copy paste from Eclipse doesn’t work.  You have to go through textedit.
  2. Postgres uses shift arrow key to select a line.  I wish Apple apps would all use the same convention.

Resources

  • http://developer.apple.com/internet/opensource/postgres.html
  • http://zanshin.net/2009/09/07/installing-postgresql-on-mac-10-6-snow-leopard/ – except for the building from source parts this was useful

Tomcat

Given how many problems I had with Postgres, I decided to follow a tutorial for Tomcat.  This one was easy to understand.  I know enough about the Mac to understand what the sudo is for (to be an admin.)  And the rest is pure UNIX – permissions, making a symbolic link, etc.

java 7 from the nyjavasig

Posted on by
2

Google java 7 a bug’s life – for approach to bug

Java 7 has come a long way since the Java road show 14 months ago when I blogged about what may or may not be in Java 7. And not always forwards. Oracle sent Donald Smith (director of project management with some coding knowledge) to the NY Java Sig.  In this blog entry, we’ll look at what features made it in, the strategy discussion from Oracle and then some details.

How does Java 7 shape up compared to the road show?

Before we get to what actually happened at the sig, let’s see how Java 7 shaped up compared to the road show.

Category Feature How to Use or Status
Modularity Project Jigsaw Java 8 – modularize the JDK to address the “Java is too big” problem.  Will not replace OSGi.
Multi-lingual Support DaVinci Machine Includes invokedynamic and more. See docs
invokeDynamic Implemented in virtual machine/bytecode to make dynamic languages faster. Invokevirtual was closest but slow because doing extra type checking. Builds CallSite so only does that extra work once and improves performance of subsequent calls. JRuby noted 20% perfromance increase.
Small languages changes (Project Coin) Diamond Operator List>Map<String, Integer> list = new ArrayList<>();
Integer Literals int num = 1_234_567;
Try Catch With Resources try (InputStream s = …) {  … }
Collection literals (like associative array) Didn’t go anywhere. Haven’t seen talked about for Java 8 either.
Other While other (undetermined) features weren’t publicized last year, they will be part of Java 8.
Performance Fork Join New APIs in Java 7.  See end of post for details.
ParallelIntArray Not in Java 7. Didn’t go anywhere.
Closures Closures/Lambda Expressions Deferred to Java 8.  Will include support for multi-core.

High Level/Strategy

Donald didn’t sound like a typical Oracle speaker. He was funny, easy to relate to and started out by talking about his biases/background It felt more personal than corporate. Usualy when Oracle prents something it sounds like legal reviewed it and stripped out a lot. The slide deck did have the standard nine line Oracle disclaimer. He did note that IBM’s Sarbanes Oxley disclaimer is twice as long as Oracle’s. People asked tough questions and he answered honestly when not knowing the answer.

Interesting things from Strategy and Q&A

  1. Timelines
    • Java 6 came out in 2006.  The four years to Java 7 was the longest time between releases anywhere.
    • Java 7 update 1 is due out in early September and will contain the major bug fixes.
    • Java 7 update 2 is due out in October and will contain garbage collection enhancements.
    • Java 8 is targeting 2012.  The speaker thinks 18 months is too soon for Java 8 because the industry is no longer used to a 2 year release cycle anymore, let alone more frequent..  He thinks we need time for the tooling to catch up.
  2. Java heath (just Java; doesn’t include JVM languages)
  3. Official terms for past are “legacy sun employee” or “legacy bea employee”.
  4. HotSpot vs JRockit
    • Survey says 70% use Sun JDK and only 5% use JRockit.
    • Oracle decided to officially kill the JRockit JVM and just use the JRockit tooling. (Mission Control, Flight Recorder and RT)
    • Over time, the two will merge with the JRockit tools being premium features.
    • Oracle emphasized performance will also be free and part of the main JDK.
    • Googlefight between the two shows Hotspot as the clear winner.
  5. JavaOne – Oracle is holding back announcements for JavaOne.  Expect some on
    • Java 7 Certifications
    • Something about the Mac.  Maybe with respect to Java 8.  (Oracle claims the delay in Java 7 on the Mac is to “make things right” with the Apple UI.)
    • Recognition that Oracle needs to clean up their name in community by announcing things at JavaOne and doing them without surprises.
  6. Open jdk is slowly becoming more open. For example, Oracle recognizes the need to open bugs to public before Oracle sees/runs triage. Looking at using jira for this.
  7. Java 7 theme is “moving Java forward.”  In other words, just get something out without waiting for all the features.

More on Java 7

The emphasis was the need to be careful about protecting the platform, the amount of work in the smallest of changes and the desire to not change the type system.  Changes fall into three categories:

  1. Language changes
    • All project coin changes noted in table up top
    • + Allowing Strings in the switch statement
    • + @SafeVarargs allowing the method itself to declare safety so all callers don’t have to suppress warnings.
    • + AutoClosable interface for I/O and JDBC 4.1 resources.  Using these within the try with resources syntax in the table above means they will get automatically closed.  It also means exceptions thrown in that auto generated finally will be suppressed but still available in the stack trace.
    • + Catching multiple unrelated exceptions catch(ExceptionType1 | ExceptionType2 e)
  2. Library changes
    • NIO – Better exceptions, more extensible to different file systems, rename behavior more consistent , more access to metadata
      • Path is the replacement for File. It understands symbolic links cross platform consistently and provides many methods.
      • Lots of methods to create/navigate/transform paths
      • Can call path.toFile() to get file from path to call old apis
      • Paths helper class to get path
      • Files helper class to copy files with lots of options such as copy with attributes or replacing exisiting attributes.  Also supports atomic move.
    • Concurrency (Fork/Join)
      • Phaser class which is similar to the cyclic barrier and countdown latch but has better synchronization and deadlock protection.  Can also add and remove threads on fly.
      • TransferQueue interface which is implmented by LinkedTransferQueue – the producer or consumer can block while waiting so dont get too far ahead
      • The key class to implement your logic in should implement RecursiveTask.  It is like RecursiveAction except that it returns a result.  All you have to do is implement the compute() method.
      • The ForkJoinPool is the executor so you can submit your task to have it run.  Methods are provided to see if it is done and get the result.  By default it uses the # available processors or you can specify explicitly.
  3. Runtime changes
    • See table up top for changes.
    • Oracle listed all the languages that can run on a JVM.  They noted that some are research projects by students and not “real” or “ready.”  I laughed because C# was on the list.  Why would you want to run C# on a JRE?
  4. Other
    • Swing nimbus look and feel is completed.  Metal is still the default.
    • Eliptic curve cryptography
    • Deadlock avoidance in classloader
    • Close method for UrlClassloader
    • Javadoc now has support for CSS.  Which means the JavaDoc now has “nice looking annoying frames”