Request for reviews (L): 6939203: JSR 292 needs method handle constants
john.r.rose at oracle.com
Wed Jun 2 16:13:25 PDT 2010
Just -Xcomp on C1 shows up the error. That would have been another problem in the nightlies.
With an assignment to 'k' my 292 unit tests passe with Xcomp and PatchALot, except for ldc of MH.
The -Xcomp test on both C1 and C2 gets a failure on a path through ciEnv::get_constant_by_index_impl, which I haven't written yet. I think I'll make a separate bug for that, so as to get the ldc stuff in circulation sooner.
On Jun 2, 2010, at 4:02 PM, Tom Rodriguez wrote:
> I suspect it just wasn't exercised, otherwise I think you would have hit this assert:
> assert(load_klass() != NULL, "klass not set");
> You can stress all C1's patching logic using -XX:+PatchALot. That forces every possible patch site to be patched.
> On Jun 2, 2010, at 3:29 PM, John Rose wrote:
>> On Jun 2, 2010, at 3:24 PM, Tom Rodriguez wrote:
>>> So shouldn't it be?
>>> + k = cc->resolve_constant(CHECK);
>> (Smacks forehead.) Right.
>> How did that ever work? I'm going to recheck my testing...
>> -- John
More information about the hotspot-compiler-dev