Crash in Compile::find_intrinsic

[John Rose]

On Jan 5, 2011, at 4:56 AM, Christian Thalinger wrote:

> I looked into this bug and the cause is that the method handles are not verified.  Running with a debug build or using -XX:+UnlockDiagnosticVMOptions -XX:+VerifyMethodHandles with a product build throws the excepted exception:

Thanks, Christian.

VerifyMethodHandles has detected that trusted JDK code has built a bad method handle.

> Caused by: java.lang.IllegalArgumentException: bad adapter (conversion=0x00000100): type mismatch: returning a sun.dyn.empty.Empty, but caller expects primitive boolean

The conversion code indicates an OP_RETYPE_RAW adapter.  The complaint is that it is adapting an Empty-valued MH to a boolean-valued one.  This is actually legal, because Empty is an "always null" type.  (See uses of is_always_null_type is methodHandles.cpp.)  More specifically, Empty is being returned because the MH itself is known to throw an exception always.  The raw adapter can retype the return type boolean (or any other type) with impunity.

So this is probably a case of false negative in VerifyMethodHandles.

> We have two problems here: (a) the conversion problem and (b) a lack of checking (either the method handles itself or in the compiler).

The conversion may be OK.  The checking logic should be examined.

-- John