[Engine-devel] JDK Failures with .gwttar

[Leonardo Bianconi]

>-----Original Message-----
>From: Itamar Heim [mailto:iheim at redhat.com]
>Sent: quarta-feira, 24 de julho de 2013 08:57
>To: Leonardo Bianconi
>Cc: Vitor de Lima; Gustavo Frederico Temple Pedrosa; engine-devel; Vojtech Szocs; Alon Bar-Lev; Juan Antonio Hernandez Fernandez
>Subject: Re: JDK Failures with .gwttar
>
>On 07/24/2013 02:46 PM, Leonardo Bianconi wrote:
>>
>>
>>> -----Original Message-----
>>> From: Itamar Heim [mailto:iheim at redhat.com]
>>> Sent: terça-feira, 23 de julho de 2013 12:44
>>> To: Leonardo Bianconi
>>> Cc: Vitor de Lima; Gustavo Frederico Temple Pedrosa; engine-devel;
>>> Vojtech Szocs; Alon Bar-Lev
>>> Subject: Re: JDK Failures with .gwttar
>>>
>>> On 07/23/2013 05:28 PM, Leonardo Bianconi wrote:
>>>> Hi Itamar!
>>>>
>>>> After rebasing our patches against the master branch from the git
>>>> repository, we noticed that the current version of the oVirt engine
>>>> uses a new version of GWT. This introduced a problem while building
>>>> the engine, which is documented in this link:
>>>>
>>>> "IBM JDK failures with .gwtar and compile" -
>>>> https://code.google.com/p/google-web-toolkit/issues/detail?id=7530
>>>>
>>>> The problem here is that we are using a different JVM (the IBM's
>>>> one) than the one that compiled the gwt files. In the beginning, we
>>>> tried to use the Open JDK, but it was too slow to be used, it takes
>>>> about 1h20m to build everything against 7m of IBM JDK, so we are using the IBM one.
>>>>
>>>> We have a question, as we are working on oVirt to work with
>>>> different architectures, should we work on fixes when we find a problem like that?
>>>> For this issue we got a workaround, using an parameter on the build
>>>> command and removing the gwt cached files.
>>>
>>> Hi Leonardo,
>>>
>>> yes, we should strive to fix these things.
>>> it would be even better if we can detect them earlier, by having a
>>> jenkins ppc slave to the ovirt jenkins, or just a local jenkins by you which will report when a patch breaks on ppc/ibm jdk for further
>consideration).
>>>
>>> questions:
>>> 1. is there a bug right now to fix? we need the build to know to add the modifier when using ibm jdk.
>> No, there is no bug in the oVirt project, it's only about the way it builds. We compiled it following the workaround:
>> "As a workaround, you can also simply remove the .gwtar files from gwt-user.jar. I think there's also a -Dgwt.usearchives=false
>system property to disable loading the gwtar files."
>>
>> The real problem is that the gwt jar contains some serializable files already compiled.  Seems gwt developers are looking for a
>solution, so we could wait a gwt solution before try to fix it.
>
>well, it failed us as well - could be this (late last night) solves it for you as well:
>http://gerrit.ovirt.org/#/c/17264/
>
>
>>>
>>> 2. (added alon) i remember we actually have an issue with ibm jdk wrt certificates or something which is worth considering in this
>>> context.
>> We didn't know about it, I will check that.
>>>
>>> 3. do you have more details on the openjdk performance issue on ppc to try and fix that one? can you open a bug on this and
>forward
>>> to us?
>> We haven't analyzed why it happens, but we know the problem is in the gwt packages, because the compilation time of the other
>packages are normal. We will compile the oVirt project with both JDKs and attach in the bug the output where it says the compilation
>time for each package, then I forward it for you.

I've opened the but for the OpenJDK performance - https://bugs.openjdk.java.net/show_bug.cgi?id=100322
Any additional info for that, you can ask me.
Regards
Leonardo Bianconi

>>>
>>> Thanks,
>>>     Itamar
>>
>> Thanks,
>> Leonardo Bianconi/Vitor de Lima/Gustavo Pedrosa
>>