298786 - Infinite loop when compiling with optimization

Status: RESOLVED FIXED

(Rhino Graveyard :: Core, defect)

Description

[Mike C.]

User-Agent:       Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.0; T312461; YComp 5.0.0.0; .NET CLR 1.1.4322; .NET CLR 2.0.50215)
Build Identifier: 

Rhino compiler gets in a infinite loop when 
compiled with -opt greather than zero, for this file: 

------------------------------­------------ 
// file infloop.js 
function infLoop(inStr) { 
  var i = -1; 
  while ((i = inStr.indexOf("%")) != -1) { 
    inStr = inStr.replace("%", ""); 
  } 

  return inStr; 
}

var str = infLoop("abc%def"); 
------------------------------­------------ 


Reproducible: Always

Steps to Reproduce:
1. save the above code in a file (infloop.js)
2. java -cp js.jar org.mozilla.javascript.tools.jsc.Main -opt 1 infloop.js

Actual Results:  
Compiler hangs, with processor at 100%.
If compiled with -opt 0 the compilation succeeds and a class file is generated.



Tested and reproduced with 1_5R5 and 1_6R1.

In 1_5R5 the loop is in the class
org.mozilla.javascript.optimizer.Optimizer
in
    private static void
    typeFlow(OptFunctionNode fn, Block theBlocks[])

I did not debug with 1_6R1 sources but I assume it is in the same place.

Comment 1

[Igor Bukanov]

Attachment: Fix

AFAICS this one-line fix is enough to fix the optimizer issue that caused the
bug. But since the bug was in Rhino since the first public release and survived
all the changes during last 5 years I am very reluctant to commit the fix
without really good testing.

Comment 2

[Igor Bukanov]

I committed the patch.

Comment 3

[Mike C.]

(In reply to comment #2)

I still have some infinite loops when I set 
Context.FEATURE_MEMBER_EXPR_AS_FUNCTION_NAME true and I use the feature, both 
in the compiler and at runtime. Should this go in a separate bug report?

Comment 4

[Igor Bukanov]

(In reply to comment #3)
> (In reply to comment #2)
> 
> I still have some infinite loops when I set 
> Context.FEATURE_MEMBER_EXPR_AS_FUNCTION_NAME true and I use the feature, both 
> in the compiler and at runtime. Should this go in a separate bug report?

Do you see this with latest CVS tip as well? It contains a few fixes in this
area. If yes, then file a bug indeed.