[OpenJDK 2D-Dev] <AWT Dev> Request for review: 8026385: [macosx] (awt) setjmp/longjmp changes the process signal mask on OS X

Phil Race philip.race at oracle.com
Tue May 13 19:59:07 UTC 2014

The JPEG code is 2d not awt ..

Apparently the root of this problem is
 > - longjmp will restore the signal mask using sigprocmask() which sets 
the signal mask for the _process_.

Hmm .. so anyone using a JNI lib that uses setjmp/longjmp would have the 
same issue.
Any place in the OS X libraries that end up using it would do the same ...
So I don't see how we can do a guaranteed fix here without knowing 
that.The change here :-

is to an upstream libpng and will likely get lost when libpng is next 
At the very least you will need to put some JDk specific comments here 
so its
blindlingly obvious when some one updates it that they are removing our

Alternatively, (and maybe better)  is there any place that libpng looks for
any customisations you need to do ?


  ifdef __APPLE__

something that we (JDK) define or the OS X compiler defines ?
ie if we asked to push this upstream would they want something different ?

Of course I have no idea if 'upstream' would want our behaviour anyway ..


On 5/13/2014 12:39 PM, David DeHaven wrote:
> JBS Issue:
> https://bugs.openjdk.java.net/browse/JDK-8026385
> Webrev:
> http://cr.openjdk.java.net/~ddehaven/8026385/jdk.0/
> Basically extends the work Staffan performed for https://bugs.openjdk.java.net/browse/JDK-8023786
> Verified by running:
> find build/macosx-x86_64-normal-server-fastdebug/jdk/objs -name '*.o' -exec nm -u {} \; | egrep '(setjmp|longjmp)'
> With this change, all references in jdk were __setjmp or __longjmp (two underscores), there are no remaining references to _setjmp/_longjmp in jdk with this change.
> -DrD-

More information about the 2d-dev mailing list