While writing unit tests for my beam pipeline using PAssert, the pipeline outputs objects fine but the test fails during comparison with following assertion error:
java.lang.AssertionError: Decode pubsub message/ParMultiDo(DecodePubSubMessage).output:
Expected: iterable with items [<PubsubMessage{message=[123, 34, 104...], attributes={messageId=2be485e4-3e53-4468-a482-a49842b87ed5, dataPipelineId=bc957aa3-17e7-46d6-bc73-0924fa5674fa, region=us-west1, ingestionTimestamp=2020-02-02T12:34:56.789Z}, messageId=null}>] in any order
but: not matched: <PubsubMessage{message=[123, 34, 104...], attributes={messageId=2be485e4-3e53-4468-a482-a49842b87ed5, dataPipelineId=bc957aa3-17e7-46d6-bc73-0924fa5674fa, region=us-west1, ingestionTimestamp=2020-02-02T12:34:56.789Z}, messageId=null}>
I also tried encapsulating expectedOutputPubSubMessage in a list (apparently original output is in an Array) to no avail. All the given PAssert examples in documentation do a simple string or keyvalue comparison.
#RunWith(PowerMockRunner.class)
public class DataDecodePipelineTest implements Serializable {
#Rule
public TestPipeline p = TestPipeline.create();
#Test
public void testPipeline(){
PubsubMessage inputPubSubMessage =
new PubsubMessage(
TEST_ENCODED_PAYLOAD.getBytes(),
new HashMap<String, String>() {
{
put(MESSAGE_ID_NAME, TEST_MESSAGE_ID);
put(DATA_PIPELINE_ID_NAME, TEST_DATA_PIPELINE_ID);
put(INGESTION_TIMESTAMP_NAME, TEST_INGESTION_TIMESTAMP);
put(REGION_NAME, TEST_REGION);
}
});
PubsubMessage expectedOutputPubSubMessage =
new PubsubMessage(
TEST_DECODED_PAYLOAD.getBytes(),
new HashMap<String, String>() {
{
put(MESSAGE_ID_NAME, TEST_MESSAGE_ID);
put(DATA_PIPELINE_ID_NAME, TEST_DATA_PIPELINE_ID);
put(INGESTION_TIMESTAMP_NAME, TEST_INGESTION_TIMESTAMP);
put(REGION_NAME, TEST_REGION);
}
});
PCollection<PubsubMessage> input =
p.apply(Create.of(Collections.singletonList(inputPubSubMessage)));
PCollection<PubsubMessage> output =
input.apply("Decode pubsub message",
ParDo.of(new DataDecodePipeline.DecodePubSubMessage()));
PAssert.that(output).containsInAnyOrder(expectedOutputPubSubMessage);
p.run().waitUntilFinish();
}
}
Apparently, someone faced the exact same issue years ago which remains unresolved. Test pipeline comparing objects using PAssert containsInAnyOrder()
Just pass the expectedOutputPubSubMessage inside an array:
PAssert
.that(output)
.containsInAnyOrder(new PubsubMessage[] { expectedOutputPubSubMessage });
The problem is that you are comparing different objects
the return of your pipeline is a PCollection and you are comparing it against PubsubMessage
you have to create a PCollection from the expectedOutputPubSubMessage
Try this:
PAssert.that(output).containsInAnyOrder(Create.of(Collections.singletonList(expectedOutputPubSubMessage));
example: https://github.com/GoogleCloudPlatform/DataflowTemplates/blob/master/src/test/java/com/google/cloud/teleport/templates/PubsubToPubsubTest.java
Related
I've been trying to make a function that returns a Map<String, Int> with the key being a certain tag and the value being the number of occurrences.
The object (simplified) from which I need to extract the info:
class Note {
List<String> tags
}
The function so far:
private fun extractTags(notes: List<Note>): Map<String, Int> {
return notes.map { note -> note.tags }
.groupBy { it }
.mapValues { it.value.count() }
}
Right now the compiler gives me a return type mismatch of Map<(Mutable)Set<String!>!, Int> and I'm not certain I'm getting the desired result (as I still can't test this properly).
I'm expecting a result something in the lines of:
(tag1, 1)
(tag2, 4)
(tag3, 14)
...
You can using Iterable#asSequence just like as Java-8 stream-api in Kotlin. then using Sequence#flatMap to merge all tags into a Sequence , and then using Sequence#groupingBy to counting each tag, for example:
private fun extractTags(notes: List<Note>): Map<String, Int> {
return notes.asSequence()
.flatMap { it.tags.asSequence() }
.groupingBy { it }.eachCount()
}
Note: both Sequence#flatMap and Sequence#groupingBy are intermediate operations, which means if the terminal operation Grouping#eachCount is not called. all of the operations on the Sequence is not ran.
While the already accepted answer unarguably solves your problem, I feel like there's a bit of an "everything looks like a nail when you have a hammer" thing going on here.
The essence of that answer is that flatMap, groupingBy, and eachCount are the methods you need to solve your problem, however, using sequences here seems completely unnecessary.
Here's the code that just operates on/with regular collections:
private fun extractTags(notes: List<Note>): Map<String, Int> {
return notes.flatMap { it.tags }
.groupingBy { it }
.eachCount()
}
I'd like to argue that this is a better solution than the one using sequences, because:
It produces the same results, since it uses the same operators.
The code is just simpler and easier to read without them.
The transformations here are simple and few, sequences get useful when you have long chains.
We are probably operating on relatively small data sets here. In my own quick measurements, the solution using sequences was about 10% faster when there are a million notes, but 17% slower when there are only ten thousand. I'll wager to guess you're closer to the latter in size of your lists. Sequences have overhead.
We aren't making use of the laziness provided by sequences at all, since we want to evaluate and return the results immediately.
You can see an excellent comparison of the two ways with pros and cons here as well for more details.
Here is your code modified to work. I changed map to flatMap. I also provided a version implemented as an extension function. Yours was failing because map> was producing a List<List<String>> where you were expecting List<String> (hence, flagMap).
fun extractTags(notes: List<Note>): Map<String, Int> {
return notes.flatMap { it.tags } // results in List<String>
.groupBy { it } // results in Pair<String, List<String>>
.mapValues { it.value.count() }
}
fun Iterable<Note>.extractTags(): Map<String, Int> {
return this.flatMap { it.tags } // results in List<String>
.groupBy { it } // results in Pair<String, List<String>>
.mapValues { it.value.count() }
}
And here is some code to test it with
import kotlin.collections.*
fun main(vararg args: String) : Unit {
var notes = ArrayList<Note>()
notes.add(Note(List<String>(1) { "tag1" }))
notes.add(Note(List<String>(4) { "tag4" }))
notes.add(Note(List<String>(14) { "tag14" }))
for((first,second) in extractTags(notes))
println("$first: $second")
for((first,second) in notes.extractTags())
println("$first: $second")
}
class Note {
constructor(strings: List<String>) {
tags = strings
}
var tags: List<String>
}
Excuse me for this late solution, but it is be the best one: as I think when you are using Kotlin you have the standard library that give you a better syntax, shorter and cleaner than the Java 8 streams.
private fun extractTags(notes: List<Note>): Map<String, Int> = notes.flatMap { it.tags }//list of String
.groupBy { it }//list of Map.Entry<String,List<String>> //List<Map.Entry<String,List<String>>>
.map {
Pair(it.key, it.value.size)
}//list of pairs(tag, count) // List<Pair(String,Int)
.toMap()//creat a map from the list of pairs
I have code like this:
//Fields
Product _prod, _existingProd;
void Test()
{
_prod = MakeAndPopulateSomeRandomProduct();
_existingProd = GetProdFromDb(1);
Mapper.CreateMap()
.AfterMap((s, d) =>
{
Console.WriteLine(d==_existingProd); //Why does this print false?
//Customize other properties on destination object
});
Mapper.Map(_prod, _existingProd);
}
When I call Test(), false is printed but I expected true. In my scenario, it is important to be able to access the original destination object via the AfterMap argument. I only included the fields to demonstrate the problem but in my real code, I don't have direct access to them. How can I access the object instances passed in to Map() when customizing the mapping?
The following example works. Probably you are using some type converter which creates new instance... Also please provide all mapping configurations to better understand the problem.
[TestFixture]
public class AfterMap_Test
{
//Fields
private Product _prod, _existingProd;
[Test]
public void Test()
{
Mapper.CreateMap<Product, Product>()
.AfterMap((s, d) =>
{
Trace.WriteLine(d == _existingProd); //Why does this print false?
//Customize other properties on destination object
});
_existingProd = new Product {P1 = "Destination"};
_prod = new Product {P1 = "Source"};
Mapper.Map(_prod, _existingProd);
}
}
internal class Product
{
public string P1 { get; set; }
}
I'm trying to verify a method call using Moq, but I can't quite get the syntax right. Currently, I've got this as my verify:
repository.Verify(x => x.ExecuteNonQuery("fav_AddFavorites", new
{
fid = 123,
inputStr = "000456"
}), Times.Once());
The code compiles, but the test fails with the error:
Expected invocation on the mock once, but was 0 times:
x => x.ExecuteNonQuery("fav_AddFavorites", new <>f__AnonymousType0<Int32, String>(123, "000456"))
No setups configured.
Performed invocations:
IRepository.ExecuteNonQuery("fav_AddFavorites", { fid = 123, inputStr = 000456 })
How can I verify the method call and match the method parameters for an anonymous type?
UPDATE
To answer the questions:
I am trying to verify both that the method was called and that the parameters are correct.
The signature of the method I'm trying to verify is:
int ExecuteNonQuery(string query, object param = null);
The setup code is simply:
repository = new Mock<IRepository>();
UPDATE 2
It looks like this is a problem with Moq and how it handles anonymous types in .Net. The code posted by Paul Matovich runs fine, however, once the code and the test are in different assemblies the test fails.
This Passes
public class Class1
{
private Class2 _Class2;
public Class1(Class2 class2)
{
_Class2 = class2;
}
public void DoSomething(string s)
{
_Class2.ExecuteNonQuery(s, new { fid = 123, inputStr = "000456" });
}
}
public class Class2
{
public virtual void ExecuteNonQuery(string s, object o)
{
}
}
/// <summary>
///A test for ExecuteNonQuery
///</summary>
[TestMethod()]
public void ExecuteNonQueryTest()
{
string testString = "Hello";
var Class2Stub = new Mock<Class2>();
Class1 target = new Class1(Class2Stub.Object);
target.DoSomething(testString);
Class2Stub.Verify(x => x.ExecuteNonQuery(testString, It.Is<object>(o => o.Equals(new { fid = 123, inputStr = "000456" }))), Times.Once());
}
##Update##
That is strange, it doesn't work in different assemblies. Someone can give us the long definition about why the object.equals from different assemblies behaves differently, but for different assemblies, this will work, any variance in the object values will return a different hash code.
Class2Stub.Verify(x => x.ExecuteNonQuery(testString, It.Is<object>(o => o.GetHashCode() == (new { fid = 123, inputStr = "000456" }).GetHashCode())), Times.Once());
One option is to "verify" it in a Callback. Obviously this needs to be done at Setup time, e.g.:
aMock.Setup(x => x.Method(It.IsAny<object>())).Callback<object>(
(p1) =>
{
dynamic o = p1;
Assert.That(o.Name, Is.EqualTo("Bilbo"));
});
None of the answers are great when your test assembly is different than the system under test's assembly (really common). Here's my solution that uses JSON serialization and then strings comparison.
Test Helper Function:
using Newtonsoft.Json;
public static class VerifyHelper
{
public static bool AreEqualObjects(object expected, object actual)
{
var expectedJson = JsonConvert.SerializeObject(expected);
var actualJson = JsonConvert.SerializeObject(actual);
return expectedJson == actualJson;
}
}
Example System Under Test:
public void DoWork(string input)
{
var obj = new { Prop1 = input };
dependency.SomeDependencyFunction(obj);
}
Example Unit Test:
var expectedObject = new { Prop1 = "foo" };
sut.DoWork("foo");
dependency.Verify(x => x.SomeDependencyFunction(It.Is<object>(y => VerifyHelper.AreEqualObjects(expectedObject, y))), Times.Once());
This solution is really simple, and I think makes the unit test easier to understand as opposed to the other answers in this thread. However, because it using simple string comparison, the test's anonymous object has to be set up exactly the same as the system under the test's anonymous object. Ergo, let's say you only cared to verify the value of a single property, but your system under test sets additional properties on the anonymous object, your unit test will need to set all those other properties (and in the same exact order) for the helper function to return true.
I created a reusable method based on Pauls answer:
object ItIsAnonymousObject(object value)
{
return It.Is<object>(o => o.GetHashCode() == value.GetHashCode());
}
...
dependency.Verify(
x => x.SomeDependencyFunction(ItIsAnonymousObject(new { Prop1 = "foo" })),
Times.Once());
Also, this can be used for property name case-insensitive comparison:
protected object ItIsAnonymousObject(object value)
{
var options = new JsonSerializerOptions { PropertyNamingPolicy = JsonNamingPolicy.CamelCase };
return It.Is<object>(o => JsonSerializer.Serialize(o, options) == JsonSerializer.Serialize(value, options));
}
Hi i need to make a VectorIterator, so i need to accept a Vector with any type. I am currently trying to define the type as * like so:
var collection:Vector.<*> = new Vector<*>()
But the compiler is complaining that the type "is not a compile time constant". i know a bug exists with the Vector class where the error reporting, reports the wrong type as missing, for example:
var collection:Vector.<Sprite> = new Vector.<Sprite>()
if Sprite was not imported, the compiler would complain that it cannot find the Vector class. I wonder if this is related?
So it looks like the answer is there is no way to implicitly cast a Vector of a type to valid super type. It must be performed explicitly with the global Vector.<> function.
So my actual problem was a mix of problems :)
It is correct to use Vector. as a generic reference to another Vector, but, it cannot be performed like this:
var spriteList:Vector.<Sprite> = new Vector.<Sprite>()
var genericList:Vector.<Object> = new Vector.<Object>()
genericList = spriteList // this will cause a type casting error
The assignment should be performed using the global Vector() function/cast like so:
var spriteList:Vector.<Sprite> = new Vector.<Sprite>()
var genericList:Vector.<Object> = new Vector.<Object>()
genericList = Vector.<Object>(spriteList)
It was a simple case of me not reading the documentation.
Below is some test code, I would expect the Vector. to cast implicitly to Vector.<*>.
public class VectorTest extends Sprite
{
public function VectorTest()
{
// works, due to <*> being strictly the same type as the collection in VectorContainer
var collection:Vector.<*> = new Vector.<String>()
// compiler complains about implicit conversion of <String> to <*>
var collection:Vector.<String> = new Vector.<String>()
collection.push("One")
collection.push("Two")
collection.push("Three")
for each (var eachNumber:String in collection)
{
trace("eachNumber: " + eachNumber)
}
var vectorContainer:VectorContainer = new VectorContainer(collection)
while(vectorContainer.hasNext())
{
trace(vectorContainer.next)
}
}
}
public class VectorContainer
{
private var _collection:Vector.<*>
private var _index:int = 0
public function VectorContainer(collection:Vector.<*>)
{
_collection = collection
}
public function hasNext():Boolean
{
return _index < _collection.length
}
public function get next():*
{
return _collection[_index++]
}
}
[Bindable]
public var selectedItems:Vector.<Category>;
public function selectionChange(items:Vector.<Object>):void
{
selectedItems = Vector.<Category>(items);
}
I believe you can refer to an untyped Vector by just calling it Vector (no .<>)
With Apache Flex 4.11.0, you can already do what you want. It might have been there since 4.9.0, but I have not tried that before.
var collection:Vector.<Object> = new Vector.<Object>()
maybe?
But i'm just speculating, haven't tried it.
var collection:Vector.<Object> = new Vector.<Object>()
but only on targeting flash player 10 cs4
I have BigDecimal objects serialized with BlazeDS to Actionscript. Once they hit Actionscript as Number objects, they have values like:
140475.32 turns into 140475.31999999999998
How do I deal with this? The problem is that if I use a NumberFormatter with precision of 2, then the value is truncated to 140475.31. Any ideas?
This is my generic solution for the problem (I have blogged about this here):
var toFixed:Function = function(number:Number, factor:int) {
return Math.round(number * factor)/factor;
}
For example:
trace(toFixed(0.12345678, 10)); //0.1
Multiply 0.12345678 by 10; that gives us 1.2345678.
When we round 1.2345678, we get 1.0,
and finally, 1.0 divided by 10 equals 0.1.
Another example:
trace(toFixed(1.7302394309234435, 10000)); //1.7302
Multiply 1.7302394309234435 by 10000; that gives us 17302.394309234435.
When we round 17302.394309234435 we get 17302,
and finally, 17302 divided by 10000 equals 1.7302.
Edit
Based on the anonymous answer below, there is a nice simplification for the parameter on the method that makes the precision much more intuitive. e.g:
var setPrecision:Function = function(number:Number, precision:int) {
precision = Math.pow(10, precision);
return Math.round(number * precision)/precision;
}
var number:Number = 10.98813311;
trace(setPrecision(number,1)); //Result is 10.9
trace(setPrecision(number,2)); //Result is 10.98
trace(setPrecision(number,3)); //Result is 10.988 and so on
N.B. I added this here just in case anyone sees this as the answer and doesn't scroll down...
Just a slight variation on Frasers Function, for anyone who is interested.
function setPrecision(number:Number, precision:int) {
precision = Math.pow(10, precision);
return (Math.round(number * precision)/precision);
}
So to use:
var number:Number = 10.98813311;
trace(setPrecision(number,1)); //Result is 10.9
trace(setPrecision(number,2)); //Result is 10.98
trace(setPrecision(number,3)); //Result is 10.988 and so on
i've used Number.toFixed(precision) in ActionScript 3 to do this: http://livedocs.adobe.com/flex/3/langref/Number.html#toFixed%28%29
it handles rounding properly and specifies the number of digits after the decimal to display - unlike Number.toPrecision() that limits the total number of digits to display regardless of the position of the decimal.
var roundDown:Number = 1.434;
// will print 1.43
trace(roundDown.toFixed(2));
var roundUp:Number = 1.436;
// will print 1.44
trace(roundUp.toFixed(2));
I converted the Java of BigDecimal to ActionScript.
We had no choices since we compute for financial application.
http://code.google.com/p/bigdecimal/
You can use property: rounding = "nearest"
In NumberFormatter, rounding have 4 values which you can choice: rounding="none|up|down|nearest". I think with your situation, you can chose rounding = "nearest".
-- chary --
I discovered that BlazeDS supports serializing Java BigDecimal objects to ActionScript Strings as well. So if you don't need the ActionScript data to be Numbers (you are not doing any math on the Flex / ActionScript side) then the String mapping works well (no rounding weirdness). See this link for the BlazeDS mapping options: http://livedocs.adobe.com/blazeds/1/blazeds_devguide/help.html?content=serialize_data_2.html
GraniteDS 2.2 has BigDecimal, BigInteger and Long implementations in ActionScript3, serialization options between Java / Flex for these types, and even code generation tools options in order to generate AS3 big numbers variables for the corresponding Java ones.
See more here: http://www.graniteds.org/confluence/display/DOC22/2.+Big+Number+Implementations.
guys, just check the solution:
protected function button1_clickHandler(event:MouseEvent):void
{
var formatter:NumberFormatter = new NumberFormatter();
formatter.precision = 2;
formatter.rounding = NumberBaseRoundType.NEAREST;
var a:Number = 14.31999999999998;
trace(formatter.format(a)); //14.32
}
I ported the IBM ICU implementation of BigDecimal for the Actionscript client. Someone else has published their nearly identical version here as a google code project. Our version adds some convenience methods for doing comparisons.
You can extend the Blaze AMF endpoint to add serialization support for BigDecimal. Please note that the code in the other answer seems incomplete, and in our experience it fails to work in production.
AMF3 assumes that duplicate objects, traits and strings are sent by reference. The object reference tables need to be kept in sync while serializing, or the client will loose sync of these tables during deserialization and start throwing class cast errors, or corrupting the data in fields that don't match, but cast ok...
Here is the corrected code:
public void writeObject(final Object o) throws IOException {
if (o instanceof BigDecimal) {
write(kObjectType);
if(!byReference(o)){ // if not previously sent
String s = ((BigDecimal)o).toString();
TraitsInfo ti = new TraitsInfo("java.math.BigDecimal",false,true,0);
writeObjectTraits(ti); // will send traits by reference
writeUTF(s);
writeObjectEnd(); // for your AmfTrace to be correctly indented
}
} else {
super.writeObject(o);
}
}
There is another way to send a typed object, which does not require Externalizable on the client. The client will set the textValue property on the object instead:
TraitsInfo ti = new TraitsInfo("java.math.BigDecimal",false,false,1);
ti.addProperty("textValue");
writeObjectTraits(ti);
writeObjectProperty("textValue",s);
In either case, your Actionscript class will need this tag:
[RemoteClass(alias="java.math.BigDecimal")]
The Actionscript class also needs a text property to match the one you chose to send that will initialize the BigDecimal value, or in the case of the Externalizable object, a couple of methods like this:
public function writeExternal(output:IDataOutput):void {
output.writeUTF(this.toString());
}
public function readExternal(input:IDataInput):void {
var s:String = input.readUTF();
setValueFromString(s);
}
This code only concerns data going from server to client. To deserialize in the other direction from client to server, we chose to extend AbstractProxy, and use a wrapper class to temporarily store the string value of the BigDecimal before the actual object is created, due to the fact that you cannot instantiate a BigDecimal and then assign the value, as the design of Blaze/LCDS expects should be the case with all objects.
Here's the proxy object to circumvent the default handling:
public class BigNumberProxy extends AbstractProxy {
public BigNumberProxy() {
this(null);
}
public BigNumberProxy(Object defaultInstance) {
super(defaultInstance);
this.setExternalizable(true);
if (defaultInstance != null)
alias = getClassName(defaultInstance);
}
protected String getClassName(Object instance) {
return((BigNumberWrapper)instance).getClassName();
}
public Object createInstance(String className) {
BigNumberWrapper w = new BigNumberWrapper();
w.setClassName(className);
return w;
}
public Object instanceComplete(Object instance) {
String desiredClassName = ((BigNumberWrapper)instance).getClassName();
if(desiredClassName.equals("java.math.BigDecimal"))
return new BigDecimal(((BigNumberWrapper)instance).stringValue);
return null;
}
public String getAlias(Object instance) {
return((BigNumberWrapper)instance).getClassName();
}
}
This statement will have to execute somewhere in your application, to tie the proxy object to the class you want to control. We use a static method:
PropertyProxyRegistry.getRegistry().register(
java.math.BigDecimal.class, new BigNumberProxy());
Our wrapper class looks like this:
public class BigNumberWrapper implements Externalizable {
String stringValue;
String className;
public void readExternal(ObjectInput arg0) throws IOException, ClassNotFoundException {
stringValue = arg0.readUTF();
}
public void writeExternal(ObjectOutput arg0) throws IOException {
arg0.writeUTF(stringValue);
}
public String getStringValue() {
return stringValue;
}
public void setStringValue(String stringValue) {
this.stringValue = stringValue;
}
public String getClassName() {
return className;
}
public void setClassName(String className) {
this.className = className;
}
}
We were able to reuse one of the available BigDecimal.as classes on the web and extended blazeds by sublassing from AMF3Output, you'll need to specify your own endpoint class in the flex xml files, in that custom endpoint you can insert your own serializer that instantiates an AMF3Output subclass.
public class EnhancedAMF3Output extends Amf3Output {
public EnhancedAMF3Output(final SerializationContext context) {
super(context);
}
public void writeObject(final Object o) throws IOException {
if (o instanceof BigDecimal) {
write(kObjectType);
writeUInt29(7); // write U290-traits-ext (first 3 bits set)
writeStringWithoutType("java.math.BigDecimal");
writeAMFString(((BigDecimal)o).toString());
} else {
super.writeObject(o);
}
}
}
as simple as that! then you have native BigDecimal support using blazeds, wooohoo!
Make sure your BigDecimal as3 class implements IExternalizable
cheers, jb
Surprisingly the round function in MS Excel gives us different values then you have presented above.
For example in Excel
Round(143,355;2) = 143,36
So my workaround for Excel round is like:
public function setPrecision(number:Number, precision:int):Number {
precision = Math.pow(10, precision);
const excelFactor : Number = 0.00000001;
number += excelFactor;
return (Math.round(number * precision)/precision);
}
If you know the precision you need beforehand, you could store the numbers scaled so that the smallest amount you need is a whole value. For example, store the numbers as cents rather than dollars.
If that's not an option, how about something like this:
function printTwoDecimals(x)
{
printWithNoDecimals(x);
print(".");
var scaled = Math.round(x * 100);
printWithNoDecimals(scaled % 100);
}
(With however you print with no decimals stuck in there.)
This won't work for really big numbers, though, because you can still lose precision.
You may vote and watch the enhancement request in the Flash PLayer Jira bug tracking system at https://bugs.adobe.com/jira/browse/FP-3315
And meanwhile use the Number.toFixed() work-around see :
(http://livedocs.adobe.com/flex/3/langref/Number.html#toFixed%28%29)
or use the open source implementations out there : (http://code.google.com/p/bigdecimal/) or (http://www.fxcomps.com/money.html)
As for the serialization efforts, well, it will be small if you use Blazeds or LCDS as they do support Java BigDecimal serialization (to String) cf. (http://livedocs.adobe.com/livecycle/es/sdkHelp/programmer/lcds/wwhelp/wwhimpl/common/html/wwhelp.htm?context=LiveDocs_Parts&file=serialize_data_3.html)
It seems more like a transport problem, the number being correct but the scale ignored. If the number has to be stored as a BigDecimal on the server you may want to convert it server side to a less ambiguous format (Number, Double, Float) before sending it.