[HARMONY-100] text/BidiWrapper issue?

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Component/s: Classlib
• Estimated Complexity: Moderate
• Assignee: Alexey Petrenko
• Reporter: Vladimir Gorr
• Created: 17/Feb/06 17:57
• Updated: 01/Nov/06 15:28
• Resolved: 01/Nov/06 15:25

Description

Let's consider the following test:

import java.text.Bidi;
public class Test {
public static void main(String[] args) throws Exception {
Bidi bd = new Bidi(new char[]

{ 's', 's', 's' }

, 0,
new byte[]

{ (byte) -7, (byte) -2, (byte) -3 }

,
0, 3, Bidi.DIRECTION_DEFAULT_LEFT_TO_RIGHT);
System.out.println("Expected 7, real " + " " + bd.getLevelAt(0));
}
}

In my opinion the JNI implementation of Bidi (text/BidiWrapper.c file, ubidi_1setPara() function) contains a potential bug, namely:
1. If the embeddingLevels argument is not NULL then _embeddingLevels variable is initialized with the JNI GetByteArrayElements() function;
2. ICU function (ubidi_setPara) initializes ICU inner structure and puts the _embeddingLevels into it;
3. If _embeddingLevels pointer is not NULL then the JNI ReleaseByteArrayElements() function (with 0 as fourth parameter) is called.
This function releases the memory (according to JNI specification) the _embeddingLevels pointer refers to;
4. After that ICU inner structure isn't initialized properly. Call of ICU ubidi_getLevels() function can return incorrect values (see java test above).

It seems the JNI_COMMIT parameter instead of "0" should be passed to the ReleaseByteArrayElements() to avoid this problem.

I'd like to underline the test mentioned above works w/o any issues for Harmony-14 contribution (although it shouldn't sometimes IMHO).
Therefore if there are any doubts in my argumentation this issue can be closed as invalid.