Unused DEBUG_CLASSFILES variable in makefiles

[Magnus Ihse Bursie]

Hi Severin and Omair,

Sorry for not replying to this earlier.

I agree that the handling of debug information in OpenJDK has not 
properly addressed the needs of the Linux distribution community. :-(

And even if this does not really help you, I can perhaps comfort you 
slightly by saying that the needs of Oracle is not properly addressed 
either. :-&

So I'd really like to spend some effort of revamping the code related to 
debug symbols, so that it properly aligns with the needs of all 
interested parties. Your mail have brought the issue higher up on the 
urgency list.

I'm trying to get a proper description of all use cases that we need to 
support. I'll get back to you when I have such a description, to check 
that I have not misunderstood the needs of the linux distribution community.

/Magnus

On 2015-02-19 10:43, Severin Gehwolf wrote:
> Hi,
>
> On Wed, 2015-02-11 at 17:52 -0500, Omair Majid wrote:
>> Hi,
>>
>> * Omair Majid <omajid at redhat.com> [2014-11-18 11:57]:
>>> * Magnus Ihse Bursie <magnus.ihse.bursie at oracle.com> [2014-10-13 07:49]:
>>>> On 2014-10-13 10:57, Erik Joelsson wrote:
>>>>> I think it's still used by the hotspot makefiles (sa-jdi.jar). Seems weird
>>>>> that we still set it in configure but then ignore the value in the jdk
>>>>> build. I would vote for resurrecting the option.
>>>> [...]
>>>>> I'm not sure if even more debug levels is the right way to go. It seems
>>>>> hard to name and that probably means it will be hard to understand.
>>>>> Perhaps completely separating native debug level and java debug level
>>>>> would make more sense?
>>>> I'm not sure "resurrecting" (which actually means adding functionality which
>>>> has been gone for quite some time now) is good, since it adds to the
>>>> complexity.
>>> Well, we do need the debug information, but I am not at all tied to
>>> specific implementation. I would be happy to try and implement something
>>> that takes a different approach than resurrecting the feature.
>>>
>>>> However, there have been recurring requests of having more control of debug
>>>> vs release builds, and I think we need to address them in some way.
>>> Any suggestions for a suitable implementation?
>> Sorry to resurrect an old thread, but this just bit me again.
> I believe it is fair to say that this issue affects *all* Linux
> distributions attempting to build from an upstream OpenJDK tree. This is
> to stress that it's a significant issue and it keeps creeping up (the
> issue being debug symbols in Java class files and natives - hotspot and
> JDK). I'm *very* interested to help find a solution to this once and for
> all.
>
> FWIW, there are a couple of checks we perform that during Fedora OpenJDK
> builds in an attempt to ensure that hotspot debug symbols are there:
> http://pkgs.fedoraproject.org/cgit/java-1.8.0-openjdk.git/tree/java-1.8.0-openjdk.spec#n1160
>
> Similarly, Java debug info is checked this way:
> http://pkgs.fedoraproject.org/cgit/java-1.8.0-openjdk.git/tree/java-1.8.0-openjdk.spec#n1177
>
> Wouldn't it be possible to add similar checks to the OpenJDK builds
> system? This is to ensure that it doesn't regress again. Thoughts?
>
> Finally, I've created this bug in order to track this:
> https://bugs.openjdk.java.net/browse/JDK-8073461
>
> Thanks,
> Severin
>