Matlab在jar包中看不到我的一些java类（不是全部）

Question

I have a problem that is driving me nuts. Matlab sees only some of my classes embeded in a JAR file

If I compile the classes outside of a package and add the path to the class in Matlab using javaaddpath, I do not encounter any problem

When I compile the class in a package and then try to access them under Matlab, I have problems. Below some Matlab code

>>javaaddpath('/Users/me/Documents/workspace/EKGTest.jar');
>>clear java
>>import com.neurosky.thinkgear.*
>>methods('EkgSense')

Methods for class com.neurosky.thinkgear.EkgSense:

EkgSense                  getClass                  notify                    reset            
addTemplate               getClassificationResults  notifyAll                  toString
equals                    hashCode                  processData               wait   

>>methods('EkgEpoch')

No methods for class EkgEpoch or no class EkgEpoch

Now, I look in the package, all of the classes are public. This is a result of jar -tf

osx:/Users/me/Documents/workspace> jar tf EKGTest.jar
META-INF/MANIFEST.MF
META-INF/REFACTORINGS.XML
com/
com/neurosky/
com/neurosky/thinkgear/
com/neurosky/thinkgear/EkgEpoch.class
com/neurosky/thinkgear/EkgEpoch.java
com/neurosky/thinkgear/EkgParameters.class
com/neurosky/thinkgear/EkgParameters.java
com/neurosky/thinkgear/EkgTemplate.class
com/neurosky/thinkgear/EkgTemplate.java
com/neurosky/thinkgear/Matlab.class
com/neurosky/thinkgear/Matlab.java
com/neurosky/thinkgear/EkgSense.class
com/neurosky/thinkgear/EkgSense.java
com/neurosky/thinkgear/DistanceArray.class
com/neurosky/thinkgear/DistanceArray.java

and below of javap --classpath

osx:/Users/me/Documents/workspace> javap -classpath /Users/me/Documents/workspace/EKGTest.jar    com.neurosky.thinkgear.EkgEpoch
Compiled from "EkgEpoch.java"
public class com.neurosky.thinkgear.EkgEpoch extends java.lang.Object implements java.lang.Cloneable{
    public int numberOfSamples;
    public float[] data;
    public com.neurosky.thinkgear.EkgEpoch(int);
    public com.neurosky.thinkgear.EkgEpoch(int, float[]);
    public com.neurosky.thinkgear.EkgEpoch(float[]);
    public com.neurosky.thinkgear.EkgEpoch(com.neurosky.thinkgear.EkgEpoch);
    public com.neurosky.thinkgear.EkgEpoch(org.json.JSONArray);
    public org.json.JSONArray toJSONArray();
    public static float[] convolve(float[], float[]);
    public float getLineNoiseAmplitude();
    public com.neurosky.thinkgear.EkgEpoch subtract(com.neurosky.thinkgear.EkgEpoch);
    public com.neurosky.thinkgear.EkgEpoch subEpoch(int, int);
    public com.neurosky.thinkgear.EkgEpoch square();
    public com.neurosky.thinkgear.EkgEpoch subtract(float);
    public com.neurosky.thinkgear.EkgEpoch diff();
    public boolean exceedValue(float, int, int);
    public com.neurosky.thinkgear.EkgEpoch smooth(int);
    public float mean();
    public float sum();
    public float max();
    public float median();
    public com.neurosky.thinkgear.EkgEpoch clone();
    public com.neurosky.thinkgear.EkgEpoch sort();
    public int[] sortIndices(com.neurosky.thinkgear.EkgEpoch);
    public float std();
    public int find_heart_beats(int[], float);
    public com.neurosky.thinkgear.EkgEpoch detrend();
    public java.lang.Object clone()       throws java.lang.CloneNotSupportedException;
}

osx:/Users/me/Documents/workspace> javap -classpath /Users/me/Documents/workspace/EKGTest.jar  com.neurosky.thinkgear.EkgSense
Compiled from "EkgSense.java"
public class com.neurosky.thinkgear.EkgSense extends java.lang.Object{
    public com.neurosky.thinkgear.EkgParameters params;
    public com.neurosky.thinkgear.EkgTemplate[] templates;
    public com.neurosky.thinkgear.EkgTemplate currentData;
    public int lastTemplateInd;
    public float lastEpochValue;
    public com.neurosky.thinkgear.EkgSense(com.neurosky.thinkgear.EkgParameters);
    public void reset();
    public void addTemplate(java.lang.String, float[][]);
    public void addTemplate(com.neurosky.thinkgear.EkgTemplate);
    public java.lang.String getClassificationResults();
    public boolean processData(float[]);
}

I am running Matlab on OSX. I have tried with Matlab 7.7.0.471 (R2008b) and 7.11.0.584 (R2010b) and got the same problem. Both Matlab are using the native OSX Java (Java 1.6.0_26-b03-384-10M3425 with Apple Inc. Java HotSpot(TM) 64-Bit Server VM mixed mode) which should be the same as the one in Eclipse (I have checked and Eclipse compiles with 1.6).

Remember that I can see the missing class no problem when I remove the package statement and some imports at the top of the java files (in all the classes of course) and when I simply add the path to the .class files (not access them in a JAR file).

Any help would be greatly appreciated. Thanks,

Jason

Answer 1

I eventually found the problem which cannot be seen above. The problem (not documented anywhere) was that some of my classes use external packages that I had not imported into Matlab. I was not planning on using any of the functionalities linked with these packages.

Nevertheless, the Matlab error message that it cannot find the class is puzzling. An error message indicating that the class cannot be used because some packages are not referenced would be most useful.

If your package uses external packages, make sure to include all of the relevant jar files in the java classpath or Matlab will not see your dependent classes.

Another tip that I found useful is that the Matlab function "import" will not return an error if you enter a package that does not exist, eg, import java.doesnotexist.* works fine. However, import java.doesnotexist.aclass will not work.

Jason

Answer 2

Another reason MATLAB will not see a class is when you compile for JRE7 and are using MATLAB 2012b (probably applies to other MATLAB releases).

The symtom is the extremly lame error

The class "JavaNuServer" is undefined.
Perhaps Java is not running.

Solution:

Compile with the javac flags

-source 1.6 -target 1.6 

Answer 3

To complement the accepted answer - there's a far better way to check if java class was loaded propertly, than through import command. Use whereisjavaclassloadingfrom function, provided by Andrew Janke in this answer .

Not only it will tell you whether the class is loaded, it will list all jars that this class is available in. It happens quite often that your class gets masked / overshadowed by another class with the same package/name, located in a different jar. That can happen eg if you try to use a newer library than one of those shipped with Matlab, and do not place it above the original one on the static java classpath.

Answer 4

This same problem can be caused by multiple types of failures. (All of them frustrating since you get NO information about what failed!) The accepted answer describes one reason and the solution. The answer by Wolfgang Kuehn gets at another possible issue, although I had trouble understanding the point so I decided to write my own answer expanding on it slightly more generally:

Each version of Matlab ships with some specific JRE that's run and supported. Different version of Matlab ship with different version of the JRE. Not matter which version of Matlab and associated JRE you have though, if you complied your external Java classes that are in the jar file with a higher (and incompatible) version of the Java compiler, Matlab will refuse to acknowledge classes even though they are in the jar. If you're lucky, you can fix this by recompiling the Java with a flag specifying compatibility with the version of Java in your instance of Matlab.

To find out which version of Java your Matlab is running, use this command at the Matlab command prompt:

version -java

Answer 5

I encountered the same problem just now. The solution is to update matlab java version by setting a new envir parameter mentioned here

https://cn.mathworks.com/matlabcentral/answers/130359-how-do-i-change-the-java-virtual-machine-jvm-that-matlab-is-using-on-windows

Matlab sucks at error message

Answer 6

One more reason why this is happening is the need of some classes to be in the static classpath of Matlab.

In this case, one of my java classes worked perfectly in Matlab2017b after adding more jars to the dynamic classpath via javaaddpath as suggested above. That prevented the java.lang.ClassNotFoundException from happening.

The same approach failed in Matlab2019a.

Now I figured that my problem is solved by simply adding my 'uber' jar containing all the jars my application needs to the classpath.txt file, like so:

   # DO NOT MODIFY THIS FILE.  IT IS AN AUTOGENERATED FILE.

/Users/andyhueni/Desktop/SPECCHIO_new_build2/SPECCHIO.app/Contents/Java/specchio-client.jar

$matlabroot/java/patch
$matlabroot/java/jar/jmi.jar
...

The location within the classpath.txt (top or end of the file) appears to be of no importance.

Modifying the classpath.txt may not be possible for all users, but there are work-arounds as listed on this excellent source of information: https://undocumentedmatlab.com/blog/static-java-classpath-hacks