[Fwd: Re: RFR (M/L): 8010722 assert: failed: heap size is too big for compressed oops]
david.holmes at oracle.com
Thu Sep 5 02:00:20 UTC 2013
On 5/09/2013 1:19 AM, Thomas Schatzl wrote:
> On Wed, 2013-09-04 at 18:22 +1000, David Holmes wrote:
>> On 4/09/2013 4:27 PM, Thomas Schatzl wrote:
>>> On Wed, 2013-09-04 at 10:59 +1000, David Holmes wrote:
>>>> On 3/09/2013 7:59 PM, Thomas Schatzl wrote:
>>>>> Hi all,
>>>>> I updated the change for 8010722 per the suggestions of David and
>>>>> Stefan and Bengt.
>>>>> - add a new OS class hook called init_ergo() to be called between
>>>>> os::init() and os::init_2().
>>>> That in itself was okay but you still call this from inside the
>>>> arguments processing code - which is what I object to.
>>> Sorry, I misunderstood then. You suggest something like:
>>> in Threads::create_vm() where the os:init() methods are called?
>> Yes - if that is feasible.
> Sure. I like the idea.
> New webrev at http://cr.openjdk.java.net/~tschatzl/8010722/webrev.5/
So you have just split parse() in two at the point where you need to
inject the os::init_large_pages. That's okay but I don't think the name
apply_ergonomics accurately reflects everything that happens in that
second function. Further, given that os::init2 is specifically defined
as being invoked after arguments have been parsed, it would now be
unclear as to whether it could come before Arguments::apply_ergonomics
(it can't!). In that light I'd prefer see very generic naming eg parse()
and parse2(), or parse_phase1(), parse_phase2().
More information about the hotspot-gc-dev