On 12/18/06, Pavel Pervov <[EMAIL PROTECTED]> wrote:
As I understand OPT pipeline, it was NOT originally designed to return
compilation error from within its internals. But now I see it being a
requirement.
This is not true. OPT pipeline is designed to return JIT_Failure if
something broken. But it's true that we can have a bug here :)
In the example in JIRA we have this stack:
harmonyvm.dll!TypeDesc::load_type_desc() Line 135 + 0x16 C++
harmonyvm.dll!type_info_get_class(TypeDesc * tih=0x095ff2f8) Line 2013 +
0x8 C++
jitrino.dll!Jitrino::DrlVMCompilationInterface::getTypeFromDrlVMTypeHandle(TypeDesc
* typeHandle=0x095ff2f8, bool isManagedPointer=false) Line 520 + 0xc C++
jitrino.dll!Jitrino::DrlVMMethodSignatureDesc::getReturnType() Line 331
C++
jitrino.dll!Jitrino::JavaByteCodeTranslator::invokeinterface(unsigned int
constPoolIndex=489, unsigned int count=1) Line 2101 + 0xd C++
jitrino.dll!Jitrino::JavaByteCodeParserCallback::parseByteCode(const
unsigned char * byteCodes=0x0470c5f8, unsigned int off=31) Line 617 + 0x36
C++
jitrino.dll!Jitrino::ByteCodeParser::parse(Jitrino::ByteCodeParserCallback
* cb=0x0012739c) Line 70 + 0x1a C++
and the body JavaByteCodeTranslator::invokeinterface has special handling
for this situation:
if (returnType == NULL) {
// This means that it was not resolved successfully but it can
be resolved later
// inside the callee (with some "magic" custom class loader for
example)
// Or respective exception will be thrown there (in the callee)
at the attempt to create (new)
// an object of unresolved type
returnType = typeManager.getNullObjectType();
}
So the method will be compiled and passed to VM.
Are you proposing to reject compilation here if some VM flag (exception
flag) is set?
--
Mikhail Fursov
