From: <he...@ca...> - 2006-05-30 00:40:44
|
At the moment, the code crashes may prevent experiments that modify the object header (in addition to GC tracing). While this discussion is important, I think it important that we not let the good be the enemy of the great by maintaining code we know will crash in existing Jikes configurations. The underlying issue is that the BumpPointer code uses the same code to handle the allocation after a reset() (in which case the bump pointer location is null) as it does to handle a bump pointer overflow. Since this occurs in the allocation fast path, I thought it better to limit the amount of code that needs to be run (as well as limit the number of changes to prevent breaking cases relying on any other coincidental correctness). I certainly have no objection to using a better solution and realize I should have explained my decisions up front. Sorry. Sincerely, Matthew Hertz Eliot Moss wrote: > Dear Matthew -- I glanced at the code and your patch, and I am not > convinced that the patch is correct, though I grasp the problem. The key > question is this: What should the allocator code do if the passed value of > "region" is 0? You skip incrementing region in addition to skipping the > "offending" stores. I am wondering if: > > (a) this routine should just return 0 if region is 0 > > (b) some caller is broken: maybe 0 shouldn't occur here? > > In either case, the patch is not the right one, and will tend to suppress a > symptom rather than fixing the underlying problem ..... > > I am posting this more broadly in hope that others more knowledgeable on > these points can also check the patch. > > Best wishes -- Eliot > > > ------------------------------------------------------- > All the advantages of Linux Managed Hosting--Without the Cost and Risk! > Fully trained technicians. The highest number of Red Hat certifications in > the hosting industry. Fanatical Support. Click to learn more > http://sel.as-us.falkag.net/sel?cmd=lnk&kid=107521&bid=248729&dat=121642 > _______________________________________________ > Jikesrvm-core mailing list > Jik...@li... > https://lists.sourceforge.net/lists/listinfo/jikesrvm-core |