Serious bug: "finally" block is run whenever "continue" is reached

[Attila Szegedi]

This isn't new to 8u40; I just tested, and it's broken in 8u31 and 8u25 too. If I had to guess, it has been like this since the initial release. I base my guess on the fact that the way Nashorn compiled finally blocks in presence of control flow transfers (break/continue/return) was unfortunately broken, I myself discovered this last December and fixed it early February – that's indeed the JIRA issue that Anthony pointed out. Unfortunately, it didn't make it in time for 8u40 :-(

If you're self-deploying, I'd suggest you get an 8u60 EA from <https://jdk8.java.net/download.html> and drop its nashorn.jar into your 8u40 JRE (it'll work). A glance at Mercurial log shows this issue with finally block inlining was picked up for 8u60-b03 build; the currently latest build is 8u60-b05, so it should be in there already. Apologies for the inconvenience. I've been doing a lot of stabilization work lately. I could get ready with some of the issues for 8u40, but there's a bunch of others that I couldn't finish in time…

Attila.

On Mar 8, 2015, at 3:22 PM, Anthony Vanelverdinghe <anthony.vanelverdinghe at gmail.com> wrote:

> Hi Tal
> 
> This might already have been solved in 9, and backported to 8u60, with bug JDK-8067139 [1]: Finally blocks inlined incorrectly
> 
> Kind regards,
> Anthony
> 
> [1] https://bugs.openjdk.java.net/browse/JDK-8067139
> 
> 
> On 8/03/2015 14:55, Tal Liron wrote:
>> This was very hard to discover, but it is real and easy to reproduce. Try running this program:
>> 
>> var finalies = 1
>> var line = 'TEST THIS LINE'
>> try {
>> while ('' !== (line = line.substring(0, line.length - 1))) {
>>  if (line.length % 2 == 0) {
>>   continue
>>  }
>>  java.lang.System.out.println(line)
>> }
>> }
>> catch (x) {
>>  java.lang.System.out.println(x)
>> }
>> finally {
>> java.lang.System.out.println('Finalies ' + (finalies++))
>> }
>> 
>> If you run this code, you'll see that the "finally" block is run every time "continue" is reached, and then the while loop continues after the "finally" block is run. Of course, the "finally" block should only be called once in this example, and the while loop should not continue, and this broken behavior can lead to serious problems.
>> 
>> (I added the "catch" block only to make sure that exceptions are not thrown, and indeed they are not. Sorry about the silly functionality of the program, but it's the result of trying to wrap my head around what was going on...)
>> 
>> Or maybe I'm crazy and am missing something trivial? The same code runs in Rhino without this problem.
>> 
>> The bug appears in the 8u40 release. :(
>> 
>> -Tal
>> 
>> 
>