RFR 8202683: Minimal VM should build cleanly on 64-bit platforms

[David Holmes]

On 7/05/2018 6:20 PM, Aleksey Shipilev wrote:
> On 05/06/2018 09:20 AM, David Holmes wrote:
>> On 5/05/2018 9:26 PM, Aleksey Shipilev wrote:
>>> RFE:
>>>     https://bugs.openjdk.java.net/browse/JDK-8202683
>>>
>>> Fix:
>>>     http://cr.openjdk.java.net/~shade/8202683/webrev.01/
>>>
>>> Minimal VM is targeted to 32-bit only, but hear me out. Recent build system changes, notably
>>> conditional GC compilation, requires build/testing with some GCs disabled. Ultimately, we want
>>> minimal VM to be buildable at all times. Today, we need to cross-compile to x86_32 to verify that,
>>> but with this change, we can build minimal on x86_64, thus simplifying our day-to-day jobs. The
>>> change itself generates jvm.cfg for minimal VM properly on both bitnesses.
>>
>> Is 64-bit client VM now able to be built routinely?
> 
> As far as I understand, the culprit before was x86_64 C1, which was fixed to allow Tiered
> compilation. Anyhow, with the build fix above, Minimal VM seems to build, bootcycle, and run fine on
> x86_64.

Okay. I don't know how well stand-alone 64-bit C1 is supported.

> 
>> I'm not sure we should be expanding the ability to build the Minimal VM.
> 
> I am thinking about the patch as removing the 32-bit special case from the jvm.cfg generation. If we
> want to forbid some JVM variants from building on some arches, we should be doing that explicitly
> somewhere else.

Perhaps. But until that happens this may give an illusion of being able 
to do more than you really can.

Anyway my unease has been noted. I don't object enough to make an issue 
of it.

Cheers,
David

> -Aleksey
>