RFR: JDK-8245032 Remove exceptions from compare.sh

Magnus Ihse Bursie magnus.ihse.bursie at oracle.com
Thu May 14 16:49:59 UTC 2020


On 2020-05-14 18:02, Erik Joelsson wrote:
> On 2020-05-14 07:23, Magnus Ihse Bursie wrote:
>> On 2020-05-14 16:05, Magnus Ihse Bursie wrote:
>>> The compare script has had a lot of exceptions for files that were 
>>> not possible to build reproducibly. Many of these has now been 
>>> fixed, and the exceptions can be removed.
>>>
>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8245032
>>> WebRev: 
>>> http://cr.openjdk.java.net/~ihse/JDK-8245032-clean-up-compare-script/webrev.01
>> This one was a bit too bold. java.base still has some issues, so we 
>> need to keep a few filters. New version, which restores some 
>> functionality:
>>
>> http://cr.openjdk.java.net/~ihse/JDK-8245032-clean-up-compare-script/webrev.02 
>>
>>
> You removed lib/classlist from the exceptions, but the comment below 
> is still there. Are you absolutely sure the classlist is still not 
> random? It's my understanding that generating this file cannot be made 
> deterministic as the exact order of the classes being compiled in the 
> JVM is timing dependent.
No, you are right. I had a private discussion with some other JDK 
developers about sorting the order of classlist when it is created, but 
that fix is not available yet. I will back that part out for now, and 
instead include it when we have a sorted classlist.

In some cases there might still lurk race conditions and nondeterminism 
that only shows up sporadically. I have run the comparison script a 
number of times on our CI system, but only time will really tell if 
there is some real issues still there.

New webrev: 
http://cr.openjdk.java.net/~ihse/JDK-8245032-clean-up-compare-script/webrev.03

/Magnus

/Magnus
>
> /Erik
>
>




More information about the build-dev mailing list