Does someone (i.e. Marcus :-) have the source to conftest.class which is used by the Java testsuite? I'm seeing an error on OS X (both JDK 1.3.1 and 1.4.1) where double values are sent as either 0.0 or Inf so I'd be interested in adding debug code to the Java side.
Could this be as simple as a struct padding or data alignment error? None of the other data types in that test fail, only double values.
The code calls a native function with some arguments to test the parameters. The reason it fails on MacOS X is that MacOS X uses a different calling convention wrt floating point arguments than the other PPC oses. The function that needs fixing is low_make_stub() function inside #ifdef HAVE_PPC_CPU and after the /* PowerOpen ABI */ comment. The function gets a bitmask specifying which (of the first 32) arguments are doubles, in case this information is needed. I haven't been able to fix the function myself, as I don't have a MacOS system to test on. If you make any changes, please make sure that it works on AIX afterwards.
/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)
Previous text:
2003-12-11 10:09: Subject: Java failure
Does someone (i.e. Marcus :-) have the source to conftest.class which is used by the Java testsuite? I'm seeing an error on OS X (both JDK 1.3.1 and 1.4.1) where double values are sent as either 0.0 or Inf so I'd be interested in adding debug code to the Java side.
Could this be as simple as a struct padding or data alignment error? None of the other data types in that test fail, only double values.
/ Jonas Walldén
Grubba has already hooked up earl-grey in Xenofarm and here's the docs that explain the ABI so now it's just a small matter of programming. :-)
http://developer.apple.com/documentation/DeveloperTools/Conceptual/MachORunt...
/ Jonas Walldén
Previous text:
2003-12-11 13:32: Subject: Java failure
The code calls a native function with some arguments to test the parameters. The reason it fails on MacOS X is that MacOS X uses a different calling convention wrt floating point arguments than the other PPC oses. The function that needs fixing is low_make_stub() function inside #ifdef HAVE_PPC_CPU and after the /* PowerOpen ABI */ comment. The function gets a bitmask specifying which (of the first 32) arguments are doubles, in case this information is needed. I haven't been able to fix the function myself, as I don't have a MacOS system to test on. If you make any changes, please make sure that it works on AIX afterwards.
/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)
Speaking of Java, there's a 7.5-specific problem in the configure.in file. The setting of java_home_guess has been modified to use a sed regexp containing (jar|zip) which unfortunately doesn't work in the seds I've tried (OS X 10.2/10.3 and Solaris 7/8):
megalon:jonasw $ java -verbose fnord.fnord.fnord 2>&1 | sed -e 's:^[[^]/]*(/.*)/[^/]*.(jar|zip)[] ].*$:\1:' -e t -e d | sed -e 's:^.* from /:/:' -e 's:/[Cc]lasses$::' -e 's:/lib$::' -e 1q megalon:jonasw $ megalon:jonasw $ java -verbose fnord.fnord.fnord 2>&1 | sed -e 's:^[[^]/]*(/.*)/[^/]*.jar[] ].*$:\1:' -e t -e d | sed -e 's:^.* from /:/:' -e 's:/[Cc]lasses$::' -e 's:/lib$::' -e 1q /i/java/j2sdk1_3_1_02/jre
Any workaround?
/ Jonas Walldén
Previous text:
2003-12-11 14:13: Subject: Java failure
Grubba has already hooked up earl-grey in Xenofarm and here's the docs that explain the ABI so now it's just a small matter of programming. :-)
http://developer.apple.com/documentation/DeveloperTools/Conceptual/MachORunt...
/ Jonas Walldén
Well, you could split it into two sed commands:
sed -e 's:^[[^]/]*(/.*)/[^/]*.jar[] ].*$:\1:' -e t -e 's:^[[^]/]*(/.*)/[^/]*.zip[] ].*$:\1:' -e t -e d
/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)
Previous text:
2003-12-12 22:24: Subject: Java failure
Speaking of Java, there's a 7.5-specific problem in the configure.in file. The setting of java_home_guess has been modified to use a sed regexp containing (jar|zip) which unfortunately doesn't work in the seds I've tried (OS X 10.2/10.3 and Solaris 7/8):
megalon:jonasw $ java -verbose fnord.fnord.fnord 2>&1 | sed -e 's:^[[^]/]*(/.*)/[^/]*.(jar|zip)[] ].*$:\1:' -e t -e d | sed -e 's:^.* from /:/:' -e 's:/[Cc]lasses$::' -e 's:/lib$::' -e 1q megalon:jonasw $ megalon:jonasw $ java -verbose fnord.fnord.fnord 2>&1 | sed -e 's:^[[^]/]*(/.*)/[^/]*.jar[] ].*$:\1:' -e t -e d | sed -e 's:^.* from /:/:' -e 's:/[Cc]lasses$::' -e 's:/lib$::' -e 1q /i/java/j2sdk1_3_1_02/jre
Any workaround?
/ Jonas Walldén
pike-devel@lists.lysator.liu.se