"Marco" == Marco Trudel <mtrudel@gmx.ch> writes:
Marco> I'm ran my private tests with my fresh gcj 4.3.
Thanks for doing this.
Marco> gcj -s -O2 --main=HelloWorld HelloWorld.java World.class -o foobar.exe
Marco> Works with gcj 4.2 but fails with gcj 4.3:
Marco> C:\DOKUME~1\Marco\LOKALE~1\Temp/ccSkeaaa.o:ccoFcaaa.jar:(.text+0x66):
Marco> undefined reference to `java::lang::String* World::getWorld()'
Marco> collect2: ld returned 1 exit status
Just FYI, often 'gcj -v' gives useful information in cases like these.
In this case, -v shows that we are not compiling World.class to a .o.
I am not sure why (I didn't investigate any deeper). The spec stuff
here is pretty awful... :(
Marco> gcj -s -O2 --main=HelloWorld HelloWorld.java World.class -ofoobar.exe
Marco> C:\DOKUME~1\Marco\LOKALE~1\Temp/ccgrbaaa.jar:0: internal compiler
Marco> error: in java_parse_file, at java/jcf-parse.c:1945
This happens because we hit a gcc_unreachable in the place where we
used to try to compile a .java file. Obviously this "shouldn't
happen" -- somehow the driver is passing the .java file to the jc1
invocation. Again, specs madness...
Marco> Well, the docs say that source and bytecode can't be mixed. So
Marco> I don't know if the docs are out of date or it's just luck that
Marco> it works with 4.2 and now it's still ok because it fails.
Ideally we would support all combinations. And, even if we don't, we
need to give a nice error rather than mysterious linker errors or
ICEs.
Please file these in bugzilla. Thanks.