RFR (XS) 8217321: [TESTBUG] utilities/test_globalDefinitions.cpp should use _LP64, not LP64

[David Holmes]

On 18/01/2019 7:31 pm, Aleksey Shipilev wrote:
> On 1/18/19 12:42 AM, David Holmes wrote:
>> On 18/01/2019 9:20 am, Aleksey Shipilev wrote:
>>> On 1/17/19 11:12 PM, David Holmes wrote:
>>>> When we build hotspot we set LP64 (whereas _LP64 presumably comes from the compiler). Shouldn't the
>>>> gtests get compiled with the same flags as the product code?
>>>
>>> Yes, I think so, and that is what happens here? Everywhere else in Hotspot we use "#ifdef _LP64",
>>> and this patch makes use of the same in gtests as well.
>>
>> I've no issue with the change, but if the same flags are used then they tests should have been
>> executed anyway. If they weren't then the same flags are not being used and perhaps they should?
> 
> I think the basic premise "LP64 is set" is incorrect.

You're correct - must be some institutional memory coming to the fore. 
The build system no longer sets it. Not even for adlc.

<snip>

>> Try ifndef LP64 - there are a few of those in ./cpu/x86/x86.ad
> 
> Oh, .ad is compiled with ADLC, which is yet another build config. I think the use of "ifndef LP64"
> there is incorrect, and it should use _LP64 as well. In fact, the rest of x86.ad uses _LP64, and
> that LP64 looks like a recent addition. And that code even fails to compile when ifndef is fixed! See:
>    https://bugs.openjdk.java.net/browse/JDK-8217371

Thanks for filing that! So glad this little discussion led to a real 
problem being discovered. :)

Cheers,
David


> -Aleksey
>