Re: Patch: Linker + Verifier fixes

[Bryce McKinlay <mckinlay at redhat dot com>]

Robert,

Could you explain what FFIBucket is for? I can't see that this has any 
effect since the closure is allocated with _Jv_Malloc - the GC will 
neither collect nor scan malloc'ed memory. I suspect the important thing 
is to make sure that the String for sym.class_name gets marked somehow, 
and I don't think that is happening with the current patch.

Ideally the closures should be collectable - if the class for which they 
are allocated were collected then the closures should as well, but since 
FFIBucket is static, it would prevent this even if the closures were 
allocated with JvAllocBytes(). It would probably be better to allocate 
the closures with JvAllocBytes and mark them as part _Jv_MarkObj when 
the class is being marked.

Casting native types to jByteArray and treating them as if they were 
Java objects isn't advised - use RawDataManaged if neccessary. Also, 
"cif" and "arg_types" shouldn't be constructed as jbyteArrays.

Cheers,

Bryce


Robert Schuster wrote:
> Hi,
> Mark Wielaard said my copyright assignment for GCC is done and I got the reply
> from the FSF's CA clerk.
>
> Tested on last time on x86 against my testsuite (see bug report) and it still
> works. :)
>
> -- What I wrote the last time (2006-01-04) still applies:
>
> I have applied all of Andrew's wishes regarding my last patch for the linker and
> verifier and fixed merge conflicts in some files.
>
> I excluded a regenerated aclocal.m4 and some Makefile.ins from the actual patch
> file to make it easier to review the interesting bits.
>
> --
>
> The original mail for this patch is here:
> http://gcc.gnu.org/ml/java-patches/2005-q4/msg00221.html
>
> and the corresponding bug report: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24616
>
> Due to some circumstances it was decided to merge two big patches into one. The
> mail about the verifier fix is here:
>
> http://gcc.gnu.org/ml/java-patches/2005-q4/msg00213.html
>
> --
>
> If there are no further requests for changes someone with write permission may
> apply it. :)
>
> --
>
> The ChangeLog:
>
> 2006-01-11  Robert Schuster  <robertschuster@fsfe.org>
>
>         * link.cc:
>         (_Jv_Linker::find_field_helper): Added checks.
>         (_Jv_Linker::find_field): Use exception swallowing class resolution
>         and added early return.
>         (_Jv_ThrowNoClassDefFoundErrorTrampoline): New function.
>         (_Jv_Linker::link_symbol_table):  Use exception swallowing class
>         resolution, added ffi_closure installation routine.
>         (_Jv_Linker::ensure_class_linked): Added string check which does
>         not trigger class resolution.
>         * java/lang/natClassLoader.cc:
>         (_Jv_FindClassNoException): New method.
>         * java/lang/Class.h:
>         (_Jv_FindClassNoException): New method declaration.
>         * include/jvm.h:
>         (_Jv_FindClassNoException): New method declaration.
>         (_Jv_FindClassFromSignatureNoException): New method declaration.
>         * prims.cc:
>         (_Jv_FindClassFromSignatureNoException): New method.
>         * gnu/gcj/runtime/FFIBucket.java: New file.
>         * gcj/javaprims.h:
>         (_Jv_equalsUtf8Classname): New method declaration.
>         (_Jv_isPrimitiveOrDerived): Dito.
>         * prims.cc:
>         (_Jv_equalsUtf8Classnames): New method.
>         (_Jv_isPrimitiveOrDerived): New method.
>         * verify.cc:
>         (ref_intersection::equals): Use new classname comparison method.
>         (type::compatible): Use new classname comparison method. Added
>         check whether LHS' type is java.lang.Object .
>         (type::resolve): Added new optional debug message and simplified
>         if-expression.
>         (type::to_array): Added codepath that generates an array type
>         without resolving the element type.
>         * sources.am: Added gnu/gcj/runtime/FFIBucket.java
>         * Makefile.in, testsuite/Makefile.in, gcj/Makefile.in, aclocal.m4:
> 	Rebuilt.