[aarch64-port-dev ] RFR(L): 8197405: Improve messages of AbstractMethodErrors and IncompatibleClassChangeErrors.

[Ningsheng Jian]

Hi,

On 2 March 2018 at 16:48, Andrew Haley <aph at redhat.com> wrote:
> On 02/03/18 01:54, Ningsheng Jian wrote:
>> Hi Goetz,
>>
>> On 28 February 2018 at 19:41, Lindenmaier, Goetz
>> <goetz.lindenmaier at sap.com> wrote:
>>> Hi,
>>>
>>> The debug build of our nightly test showed an issue of our latest builds.
>>> Thus new webrev with change in throw_AbstractMethodErrorWithMethod()
>>> http://cr.openjdk.java.net/~goetz/wr18/8197405-ameExMsg/webrev.05/
>>>
>>> Besides the one problem in my test which is fixed by the new webrev,
>>> all nightly tests were green.
>>>
>>
>> I am not a reviewer. Just eyeballed the aarch64 changes:
>>
>> src/hotspot/cpu/aarch64/vtableStubs_aarch64.cpp:
>>
>> 209   __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub());
>>
>> Another ")" is missing at the end?
>>
>> Compared to other ports, the aarch64 invokeinterface should also be
>> updated. Since you've disabled the tests on aarch64, this can also be
>> fixed in another patch. CCed aarch64 people.
>
> Hi,
>
> Since you've already reviewed this, can you please fix it?
>

I think adding the following patch should work for AArch64:

--------
--- a/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp
@@ -3440,6 +3440,8 @@ void TemplateTable::invokeinterface(int byte_no) {

   Label no_such_interface, no_such_method;

+  // Preserve method for throw_AbstractMethodErrorVerbose.
+  __ mov(r16, rmethod);
   // Receiver subtype check against REFC.
   // Superklass in r0. Subklass in r3. Blows rscratch2, r13
   __ lookup_interface_method(// inputs: rec. class, interface, itable index
@@ -3460,8 +3462,10 @@ void TemplateTable::invokeinterface(int byte_no) {
   __ subw(rmethod, rmethod, Method::itable_index_max);
   __ negw(rmethod, rmethod);

+  // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
+  __ mov(rlocals, r3);
   __ lookup_interface_method(// inputs: rec. class, interface, itable index
-                             r3, r0, rmethod,
+                             rlocals, r0, rmethod,
                              // outputs: method, scan temp. reg
                              rmethod, r13,
                              no_such_interface);
@@ -3490,7 +3494,8 @@ void TemplateTable::invokeinterface(int byte_no) {
   // throw exception
   __ restore_bcp();      // bcp must be correct for exception handler
  (was destroyed)
   __ restore_locals();   // make sure locals pointer is correct as
well (was destroyed)
-  __ call_VM(noreg, CAST_FROM_FN_PTR(address,
InterpreterRuntime::throw_AbstractMethodError));
+  // Pass arguments for generating a verbose error message.
+  __ call_VM(noreg, CAST_FROM_FN_PTR(address,
InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
   // the call_VM checks for exception, so we should never return here.
   __ should_not_reach_here();

@@ -3498,8 +3503,9 @@ void TemplateTable::invokeinterface(int byte_no) {
   // throw exception
   __ restore_bcp();      // bcp must be correct for exception handler
  (was destroyed)
   __ restore_locals();   // make sure locals pointer is correct as
well (was destroyed)
+  // Pass arguments for generating a verbose error message.
   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
-                   InterpreterRuntime::throw_IncompatibleClassChangeError));
+
InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3,
r0);
   // the call_VM checks for exception, so we should never return here.
--------

I enabled Goetz's newly added two cases and they passed on my AArch64
platform. More regression tests are still running.

Thanks,
Ningsheng