The fun thing is that it seems that they are aware of the problem and in fact they’re trying to use a workaround internally in the chromium source code.
For instance, this is the first workaround I found:
https://codereview.chromium.org/180273005/patch/80001/90003

which is basically looking into the “/data/app-lib/{packagename}-1/” folder for the missing libraries.
And in fact, inside the comments of that file you can read this:

Native library directory in an updated package is a symbolic link
to a directory in /data/app-lib/, for example:
/data/data/com.android.chrome/lib -> /data/app-lib/com.android.chrome[-1].
When updating the application, the PackageManager create a new directory,
e.g., /data/app-lib/com.android.chrome-2, and remove the old symlink and
recreate one to the new directory. However, on some devices (e.g. Sony Xperia),
the old directory was deleted, but deleting the old symlink failed,
and that makes system loading library from the /system/lib directory.

Then I noticed another approach which is probably better: http://goo.gl/cbCAmC

which is basically doing what the PackageManager was supposed to do, unzip the shared libraries from the apk file to the “lib” dir of the application itself.

Btw, I’m gonna try both of them and verify if they actually work so I can finally stop seeing that annoying bug in the Google Play Console!

After googling for my problem I found that it was because a bug in the LocalActivityManager class: http://code.google.com/p/android/issues/detail?id=12359
This is exactly my problem! In fact, debugging the android code ( see this post for debugging android code ) at that point it was clear that the record it’s not removed from the internal mActivities map.

Fortunately there are Reflections which can help me! I added this piece of code after the call to the destroy method:


Field mActivitiesField = getLocalActivityManager().getClass().getDeclaredField( "mActivities" );
mActivitiesField.setAccessible( true );
((Map) mActivitiesField.get( getLocalActivityManager() )).remove( id );


and that trick does the job

package
{
   import flash.display.Sprite;
   public class ternary_test extends Sprite
   {
      public function ternary_test()
      {
         var n: Number = 0;
         var d: int = 3;
         var k: int = 5;
         var j: int = 10;
         n = 1 + ( -( d == 0 ? k : j ) );
         trace( "n=" + n );
      }
   }
}

will cause the compiler to crash with the following stack trace:

compile:

    [mxmlc] Loading configuration file /Applications/Adobe Flash Builder 4 Plug-in/sdks/3.5.0/frameworks/flex-config.xml

    [mxmlc] Error: null

    [mxmlc] java.lang.NullPointerException

    [mxmlc] 	at macromedia.asc.semantics.ConstantEvaluator.evaluate(ConstantEvaluator.java:1290)

    [mxmlc] 	at macromedia.asc.parser.UnaryExpressionNode.evaluate(UnaryExpressionNode.java:51)

    [mxmlc] 	at macromedia.asc.semantics.ConstantEvaluator.evaluate(ConstantEvaluator.java:2153)

    [mxmlc] 	at macromedia.asc.parser.ListNode.evaluate(ListNode.java:44)

    [mxmlc] 	at macromedia.asc.semantics.ConstantEvaluator.evaluate(ConstantEvaluator.java:1497)

    [mxmlc] 	at macromedia.asc.parser.BinaryExpressionNode.evaluate(BinaryExpressionNode.java:56)

    [mxmlc] 	at macromedia.asc.semantics.ConstantEvaluator.evaluate(ConstantEvaluator.java:2124)

    [mxmlc] 	at macromedia.asc.parser.ArgumentListNode.evaluate(ArgumentListNode.java:45)...

… items — An optional list of one or more comma-separated values, **or a Vector**, to insert into the Vector at the position specified in the startIndex parameter.


var v1: Vector. = new Vector.();
v1.push(1);
v1.push(2);
v1.push(3);
v1.push(4);

// Expected result: [1,100,200,3,4]
// Actual result: [1,0,3,4]

It seems that the third parameter is being casted into int and so only ‘0’ has been inserted into the first Vector!

I tried to look into Adobe jira bugbase and what I’ve found there is that it was a known bug since long time and it has been also moved to the tamarin bugbase.

The problem is that at the end they simply closed the bug as invalid bug because the documentation was wrong!

Funny

Today while surfing the net I received a notification of an update of my Java installation to the Update 10 in order to see correctly the contents of that site.. bla bla bla..

Sometime later I switched to Flex and at the first execution of my ant script for compiling a flex project I suddenly received this beautiful error:
[exec] Error: java/security/MessageDigest
[exec] java.lang.NoClassDefFoundError: java/security/MessageDigest
[exec] at java.io.ObjectStreamClass.computeDefaultSUID(Unknown Source)
[exec] at java.io.ObjectStreamClass.access$100(Unknown Source)
[exec] at java.io.ObjectStreamClass$1.run(Unknown Source)
[exec] at java.security.AccessController.doPrivileged(Native Method)
[exec] at java.io.ObjectStreamClass.getSerialVersionUID(Unknown Source)
[exec] at java.io.ObjectStreamClass.initNonProxy(Unknown Source)
[exec] at java.io.ObjectInputStream.readNonProxyDesc(Unknown Source)
[exec] at java.io.ObjectInputStream.readClassDesc(Unknown Source)
[exec] at java.io.ObjectInputStream.readNonProxyDesc(Unknown Source)
[exec] at java.io.ObjectInputStream.readClassDesc(Unknown Source)
[exec] at java.io.ObjectInputStream.readOrdinaryObject(Unknown Source)
[exec] at java.io.ObjectInputStream.readObject0(Unknown Source)
[exec] at java.io.ObjectInputStream.readObject(Unknown Source)
[exec] at flex.util.SerializedTemplateFactory.load(SerializedTemplateFactory.java:30)
[exec] at flex2.compiler.util.VelocityManager.getTemplate(VelocityManager.java:162)
[exec] at flex2.compiler.util.VelocityManager.getTemplate(VelocityManager.java:151)
[exec] at flex2.compiler.util.VelocityManager.getTemplate(VelocityManager.java:137)
[exec] at flex2.compiler.as3.genext.GenerativeSecondPassEvaluator.generateSupportCode(GenerativeSecondPassEvaluator.java:256)
[exec] at flex2.compiler.as3.genext.GenerativeSecondPassEvaluator.modifySyntaxTree(GenerativeSecondPassEvaluator.java:355)
[exec] at flex2.compiler.as3.binding.BindableSecondPassEvaluator.evaluate(BindableSecondPassEvaluator.java:102)
[exec] at macromedia.asc.parser.ClassDefinitionNode.evaluate(ClassDefinitionNode.java:106)
[exec] at flash.swf.tools.as3.EvaluatorAdapter.evaluate(EvaluatorAdapter.java:338)
[exec] at macromedia.asc.parser.StatementListNode.evaluate(StatementListNode.java:60)
[exec] at flash.swf.tools.as3.EvaluatorAdapter.evaluate(EvaluatorAdapter.java:923)
[exec] at macromedia.asc.parser.ProgramNode.evaluate(ProgramNode.java:80)
[exec] at flex2.compiler.as3.genext.GenerativeExtension.parse2(GenerativeExtension.java:114)
[exec] at flex2.compiler.as3.Compiler.parse2(Compiler.java:386)
[exec] at flex2.compiler.API.parse2(API.java:2390)
[exec] at flex2.compiler.API.parse2(API.java:2348)
[exec] at flex2.compiler.API.batch2(API.java:375)
[exec] at flex2.compiler.API.batch(API.java:1117)
[exec] at flex2.compiler.API.compile(API.java:1290)
[exec] at flex2.compiler.API.compile(API.java:1210)
[exec] at flex2.tools.Compc.compc(Compc.java:158)
[exec] at flex2.tools.Compc.main(Compc.java:41)
so I tried also to compile another project before go crazy, but I got this another java funny error, which was more or less unreadable as the previous one!

Fortunately I remembered the update I did of Java! And in fact after remove that update everything worked again!

The problem with the tool is because in order to find the correct bounds of any element on the stage it use the .concatenatedMatrix property of the element itself ( which represents the combined transformation matrixes of the display object and all of its parent objects, back to the root level ).
BUT if any of its parent display object ( not the root itself ) has the cacheAsBitmap property set to true then the concatenatedMatrix returns a wrong value!

I put a real simple flash example of what I mean. Just click on any of the 2 colored objects to activate the transform tool, then click on the “cache” button and click again on those objects to see the difference.

[kml_flashembed publishmethod=”static” fversion=”10.0.22″ movie=”https://blog.sephiroth.it/wp-content/uploads/2008/03/cacheasbitmap_hell.swf” width=”450″ height=”450″ targetclass=”flashmovie” bgColor=”#ffffcc”]

[/kml_flashembed]

P.S. I’ve reported the bug in the flash player bugbase, if anyone is interested in watch or vote for it.

[kml_flashembed publishmethod=”static” fversion=”10.0.22″ movie=”https://blog.sephiroth.it/wp-content/uploads/2007/11/canvas_bug.swf” width=”250″ height=”290″ targetclass=”flashmovie”]

[/kml_flashembed]

And this is the code used:


< ?xml version="1.0" encoding="utf-8"?>

1











As you can see once the image is scaled and it’s bigger that its parent, you wont be able to see it top left corner anymore. This is an annoying bug for me!

Later I found that someone opened a ticket in the Flex bug management system here: http://bugs.adobe.com/jira/browse/SDK-13009
A couple of days ago I received a notification that the bug has been closed… and incredible they closed it as NOT A BUG!
I can understand that finding the bug in a file with 5000 and more lines of code it’s not easy and can cause horrible headache, but the reason they gave it’s nosense! It’s completely a different thing that the reason of this bug.

Moreover what’s the connection between flash and html?

Flex have been designed since the 1.0 version as a framework for building RIA (Rich Internet Applications).
A RIA have to behave as a desktop application that runs using the web browser. One of the advantages of Flex agains Ajax is that the first technology provide to the user a behaviour which is much more similar to desktop applications that the one provided using HTML+Javascript. And so my questions is: why should Flex behave like HTML when one of its most important features is to be different from HTML ?
Flex behaves like other SDKs used to develop desktop apps in many situations which are much less important than the one hilighted above. So why close this bug (yes. it is a bug) providing as explanation that it is correct because it behaves like HTML ?

But the interesting thing I found was another!
Flash does not read the sharedojects correctly (at least the amf0 ones)!

To reproduce this bug follow these steps:

Create a new document (flash8 / as2)
– Add this code in the timeline:
var so:SharedObject = SharedObject.getLocal("flash8", "/");
so.clear();
so.data.obj1 = new XML("");
so.data.obj2 = new Date();
so.data.obj3 = so.data.obj2;
so.data.obj4 = {a:1, b:so.data.obj1, c:[1,2,3], d:so.data.obj2};
so.data.obj5 = so.data.obj4.c;
so.flush();
you will see this in the debug view variables:
Variabile _level0.so = [oggetto #1, classe 'SharedObject'] {
data:[oggetto #2, classe 'Object'] {
obj1:[oggetto #3] {
;
},
obj2:[oggetto #4, classe 'Date'] {Sat Sep 22 19:31:31 GMT+0200 2007},
obj3:[oggetto #4, classe 'Date'],
obj4:[oggetto #5, classe 'Object'] {
d:[oggetto #4, classe 'Date'],
c:[oggetto #6, classe 'Array'] [
0:1,
1:2,
2:3
],
b:[oggetto #3],
a:1
},
obj5:[oggetto #6, classe 'Array']
}
}
ok, now remove all the code you just written and leave just this line:
var so:SharedObject = SharedObject.getLocal("flash8", "/");
and see the variables list again, you will see this:
Variabile _level0.so = [oggetto #1, classe 'SharedObject'] {
data:[oggetto #2, classe 'Object'] {
obj1:[oggetto #3] {
;
},
obj2:[oggetto #4, classe 'Date'] {Sat Sep 22 19:31:31 GMT+0200 2007},
obj3:[oggetto #5, classe 'Date'] {Sat Sep 22 19:31:31 GMT+0200 2007},
obj4:[oggetto #6, classe 'Object'] {
d:[oggetto #7, classe 'Date'] {Sat Sep 22 19:31:31 GMT+0200 2007},
c:[oggetto #8, classe 'Array'] [
0:1,
1:2,
2:3
],
b:[oggetto #6, classe 'Object'],
a:1
},
obj5:non definito
}
}
they are different! (both data.obj5 and data.obj4.b changed)

For what I understand during my experiments it seems that flash does not write into its internal references table the XML objects while reading back the file and so next time it open the sharedobject all the references after the xml are broken or just they point to a different object.

Read the full post: http://aprogrammerdiary.com/../bug_in_bitmapdata.html

Flex 2.0.1 SDK hotfix 1 (build 159086) is a collection of bug fixes that have been completed since the Flex 2.0.1 release on January 5, 2007. This hotfix release is being made available to Flex customers who are experiencing any of the issues described in this document and require the provided bug fixes. Installation of this hotfix release is not recommended for customers who are not experiencing one of the described issues.

BTW, it seems they haven’t fixed the Nullpointer exception caused by invalid xmllists..