Bug 49278 - Branch target offset too large for short
Summary: Branch target offset too large for short
Status: NEW
Alias: None
Product: BCEL - Now in Jira
Classification: Unclassified
Component: Main (show other bugs)
Version: unspecified
Hardware: PC Windows XP
: P2 normal
Target Milestone: ---
Assignee: issues@commons.apache.org
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2010-05-12 10:39 UTC by cfeldmann
Modified: 2010-05-12 10:39 UTC (History)
0 users



Attachments
Fix to avoid "Branch target offset too large for short" exception. (2.78 KB, application/octet-stream)
2010-05-12 10:39 UTC, cfeldmann
Details

Note You need to log in before you can comment on or make changes to this bug.
Description cfeldmann 2010-05-12 10:39:03 UTC
Created attachment 25432 [details]
Fix to avoid "Branch target offset too large for short" exception.

When BCEL encounters a BranchInstruction offset that is >= 32767, it throws the following exception:

com.sitraka.pas.sandbox.org.apache.bcel.generic.ClassGenException: Branch target offset too large for short
at com.sitraka.pas.sandbox.org.apache.bcel.generic.BranchInstruction.dump(BranchInstruction.java:99)
at com.sitraka.pas.sandbox.org.apache.bcel.generic.InstructionList.getByteCode(InstructionList.java:993)
at com.sitraka.pas.sandbox.org.apache.bcel.generic.MethodGen.getMethod(MethodGen.java:616)
..........

Bug 35405 was previously filed about this exception and closed because the class that was causing the error and the suggested solution were both violating the rules of the language. My situation is a little different, so I am filing a new bug. 

I am inserting bytecodes into a method that is large, but within the limits defined by the language. Due to the size of the method, a number of the branch targets are large, but still small enough to fit in a short. After I insert bytecodes, the method size is still smaller than the method size limit but the large branch targets increase enough so that they no longer fit in a short and the exception occurs. 

I have solved the problem and am including a patch. My solution is to check the InstructionList for any branch targets that are too large to fit in a short and append a goto and a goto_w after every such instruction. I set the goto's target to the instruction that followed the branch instruction before I did my append so that the default behaviour is to maintain the original path of execution and jump over the goto_w. I set the goto_w's target to the branch instruction's target, then I set the branch instruction's target to the goto_w. I am doing this only for IfInstructions because JSRs, GOTOs and Selects seem to handle large branch targets in their implementations of updatePosition. This works and is within the rules defined by the language as far as I can tell.