<AWT Dev> [OpenJDK 2D-Dev] Fwd: openjdk7 - use installed png and jpeg libraries

Dmitry Samersoff dmitry.samersoff at oracle.com
Mon Dec 24 10:58:40 PST 2012


Phil,

Thanks!

Could someone from java2d team file a bug to have it fixed?

-Dmitry

On 2012-12-24 22:51, Phil Race wrote:
> On 12/24/2012 10:45 AM, Dmitry Samersoff wrote:
>> Phil,
>>
>> Thank you for the explanation.
>>
>> My main concern is a mismatch between jpeg library and header.
>>
>> i.e. if we use bundled libjpeg could we use bundled header as well.
> 
> Do you mean "must we" ? I think that has to be a yes - they need to match.
> I don't see the diff but I'd think the first priority would be to make
> the bundled
> jpeg being used pick up its own headers during compilation.
> 
> -phil.
> 
>> -Dmitry
>>
>>
>> On 2012-12-24 22:42, Phil Race wrote:
>>> libpng and giflib are used solely by splashscreen so I don't think that
>>> matters so much.
>>> Distro builds can do this fairly safely since they link against a known
>>> quantity.
>>> Perhaps not so much the Oracle JDK which builds on one distro but is run
>>> on many ..
>>>
>>> But jpeg might matter.
>>> Oracle JDK does not/cannot use the system libjpeg for all purposes
>>> because of some necessary proprietary code in the version used by
>>> ImageIO.
>>> So a change would have to be confined to the OpenJDK and even if done
>>> solely for splashscreen in Oracle JDK would mean two libjpegs being
>>> loaded ..
>>>
>>> BTW I responded on the thread you cite and mentioned this
>>> http://mail.openjdk.java.net/pipermail/2d-dev/2012-May/002509.html
>>> I do not believe that switching alone would fix anything here.
>>>
>>> I'd prefer at this point to defer such a change for jpeg to early in
>>> JDK 9.
>>>
>>> -phil.
>>>
>>> On 12/24/2012 2:10 AM, Damon Hart-Davis wrote:
>>>> If done right then this might help deal with the worrying
>>>> incompatibility I noted before between the colour space values/range
>>>> between OpenJDK and Oracle JDK:
>>>>
>>>> http://mail.openjdk.java.net/pipermail/2d-dev/2012-May/002496.html
>>>>
>>>> Rgds
>>>>
>>>> Damon
>>>>
>>>>
>>>> On 24 Dec 2012, at 09:52, Anthony Petrov wrote:
>>>>
>>>>> (CC'ing awt-dev@ and 2d-dev@)
>>>>>
>>>>> There's an RFE:
>>>>>
>>>>> 6565114: Allow using system copies of libjpeg, libpng and giflib for
>>>>> splashscreen
>>>>> http://bugs.sun.com/view_bug.do?bug_id=6565114
>>>>>
>>>>> I think that the scope of this RFE might be extended to support using
>>>>> system copies of the libraries for the core JDK as well, not only for
>>>>> the splashscreen.
>>>>>
>>>>> However, the RFE is currently closed as a "Future Project" because
>>>>> the contributed patch contains some issues and the contributor was
>>>>> not available for a while. See the review thread for details (a link
>>>>> is in the Description of the bug).
>>>>>
>>>>> In any case, if you plan to integrate such a change to OpenJDK 8, it
>>>>> should be reviewed here on the awt-dev@ and 2d-dev@ mailing lists.
>>>>>
>>>>> -- 
>>>>> best regards,
>>>>> Anthony
>>>>>
>>>>> On 12/22/2012 11:15, Dmitry Samersoff wrote:
>>>>>> Anthony,
>>>>>> Should we do it for all system or at least,
>>>>>> use bundled header for bundled library?
>>>>>> -Dmitry
>>>>>> -------- Original Message --------
>>>>>> Subject: openjdk7 - use installed png and jpeg libraries
>>>>>> Date: Fri, 21 Dec 2012 22:24:02 -0500
>>>>>> From: Mikhail T.<mi+thun at aldan.algebra.com>
>>>>>> To: glewis at FreeBSD.org
>>>>>> CC: java at FreeBSD.org
>>>>>> Hello!
>>>>>> The attached diff teaches java/openjdk7 to rely on the
>>>>>> graphics/{png,jpeg} ports instead of compiling libpng and libjpeg
>>>>>> from
>>>>>> sources bundled by Oracle.
>>>>>> This is especially important for jpeg, because currently build
>>>>>> uses the
>>>>>> already installed headers while compiling the bundled sources -- in
>>>>>> case
>>>>>> of any discrepancies there may be nasty problems.
>>>>>> It is also just cleaner and affords the use of better optimized image
>>>>>> libraries (such as assembly-optimized png and jpeg-turbo).
>>>>>> Unfortunately, the port currently does not include
>>>>>> check/regression-test
>>>>>> target, so I my limited testing consisted of simply restarting
>>>>>> vuze --
>>>>>> this worked, but I am not certain, it exercises the modified code.
>>>>>> Please, take a look. Thank you! Yours,
>>>>>>      -mi
>>
> 


-- 
Dmitry Samersoff
Oracle Java development team, Saint Petersburg, Russia
* Give Rabbit time, and he'll always get the answer



More information about the awt-dev mailing list