[
https://issues.apache.org/jira/browse/BCEL-195?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14711890#comment-14711890
]
Mark Roberts commented on BCEL-195:
-----------------------------------
There is no space/performance hit because they were never unique. They are
not defined in InstructionConstants so Instruction.readInstruction always
creates a new instance for every branch instruction (among others). I guess I
hadn't made that clear before. The only purpose of my change to
InstructionComparator is to trick the current targeter design into working the
way I think it should (via tricking HashSet).
Daikon does branch instruction replacement all the time - but that would be the
opposite of a very basic test case. :-)
> addition of hashCode() to generic/Instruction.java breaks Targeters
> -------------------------------------------------------------------
>
> Key: BCEL-195
> URL: https://issues.apache.org/jira/browse/BCEL-195
> Project: Commons BCEL
> Issue Type: Bug
> Components: Main
> Affects Versions: 6.0
> Reporter: Mark Roberts
> Attachments: bcel195.diff, targeters.diff
>
>
> [Revision 1532198|http://svn.apache.org/r1532198] added a {{hashCode()}}
> function to the Instruction class. Unfortunately, this breaks the
> Instruction targeting mechanism. I understand the goal of trying to reuse
> instructions - an 'iadd' is the same as any other 'iadd'. However, one
> 'goto 50' is not the same as another 'goto 50' due to the way Targeters are
> implemented. If branch instructions are reused, then only one entry gets put
> on the Targeter list. So when some api is used to modify the instruction
> list and location 50 becomes location 52 ONLY ONE of the branches gets
> updated. A very bad thing. So unless you modify the hash to special case
> branch instructions (and there might be other instructions needing special
> treatment as well) its broken. We fixed it by simply commenting the hash out
> to make things like they used to be and all works great.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
