RFR (M): 8139595: MethodHandles::remove_dependent_nmethod is not MT safe
john.r.rose at oracle.com
Thu Nov 12 20:53:33 UTC 2015
On Nov 11, 2015, at 7:43 AM, Vladimir Ivanov <vladimir.x.ivanov at oracle.com> wrote:
> Though previous comments were focused on dependency management code refactoring, I'd like to hear opinions about the fix itself - dependency management changes for j.l.i.CallSites to avoid modifications during concurrent traversals by GC threads.
Indeed, that's the tough part.
I'd like to ask a favor: Since the refactoring moves a large block of code, could you please split out the simple textual move into a separate, prior webrev? That way I can see exactly what tweaks were made during the move to the new file (instKlass => new depCon).
(This is a general feature of move-refactorings: The webrev reports a large deletion in one place and a large, similar insertion in another place. The reviewer asks, "What changed during the move?", but the webrev doesn't help. The only way to view any small textual changes is to first discount all of the effects of the block move, and *then* look at the changes to the text block as if they were post-move tweaks. The block move is a trivial change, but it should be isolated in another webrev to avoid interfering with the main review.)
More information about the hotspot-dev