RFR: JDK-8162354: Unable to build JDK 9 on a Sparc T7-1 with default boot-jdk 8.0

[Erik Joelsson]

Maybe, but the problem is that we don't know what the next chip name 
will be. The last generation was T5, so the letter may also change. I 
think it's safer to go with what we know at this point.

/Erik


On 2016-08-03 13:26, Seán Coffey wrote:
> One suggestion to future proof your fix might be to pattern match 
> against any CPU of M7 of greater perhaps. Something like SPARC-M[7+] 
> perhaps.
>
> Regards,
> Sean.
>
> On 03/08/16 11:47, Erik Joelsson wrote:
>> Followup on this. I did a compare build on Solaris with 8GA and 8u20. 
>> There is a slight difference for these class files:
>>
>> /jdk.compiler/com/sun/tools/javac/resources/CompilerProperties$Errors.class 
>>
>> /jdk.compiler/com/sun/tools/javac/resources/CompilerProperties$Fragments.class 
>>
>> /jdk.compiler/com/sun/tools/javac/resources/CompilerProperties$Notes.class 
>>
>> /jdk.compiler/com/sun/tools/javac/resources/CompilerProperties$Warnings.class 
>>
>> /jdk.compiler/com/sun/tools/javac/resources/CompilerProperties.class
>>
>> The source file is generated and also differs. Without digging 
>> deeper, it looks like it's just ordering of methods, fields and/or 
>> inner classes, which is caused by the generator program not 
>> specifying the order, likely by using an unordered collection to 
>> store them during generation. While this is most likely fine, it's 
>> good practice to have the build more deterministic.
>>
>> /Erik
>>
>>
>> On 2016-08-03 11:11, Erik Joelsson wrote:
>>> New webrev: http://cr.openjdk.java.net/~erikj/8162354/webrev.02/
>>>
>>> I found that the first working update was 8u20 and changed to use 
>>> that instead for minimal impact.
>>>
>>> /Erik
>>>
>>> On 2016-08-03 09:08, Erik Joelsson wrote:
>>>>
>>>>
>>>> On 2016-08-03 04:20, David Holmes wrote:
>>>>> Hi Erik,
>>>>>
>>>>> On 3/08/2016 1:11 AM, Erik Joelsson wrote:
>>>>>> Hello,
>>>>>>
>>>>>> The default bootjdk for 9 is JDK 8. Unfortunately JDK 8 does not 
>>>>>> work on
>>>>>> newer M7 sparc hardware. This has of course been fixed in an update.
>>>>>> This patch changes the Jib profile configuration to use 8u102 as 
>>>>>> bootjdk
>>>>>> on Solaris sparc if the cpu is of the M7 type.
>>>>>
>>>>> Isn't 8u102 a bit too recent to use - shouldn't we use the oldest 
>>>>> version that works on M7?
>>>>>
>>>> That is a good point. Either go with latest greatest when changing 
>>>> or try to find the minimum change. We had a similar problem in JDK 
>>>> 8 where 7GA did not work on macosx. At that time we took the first 
>>>> working version. I will see if I can figure out which one it is at 
>>>> least.
>>>>> Does this actually affect official builds (do they use M7) or does 
>>>>> this simply enable use of M7 in the other build processes (JPRT etc)?
>>>>>
>>>> Unclear. The new hardware we received is M7, so without a change we 
>>>> can't use that hardware for any builds, which would be sad since 
>>>> the machine is very fast. I suspect that down the line we will want 
>>>> to use this class of hardware in all types of build scenarios. IMO, 
>>>> we get adequate testing that the 8GA as boot (which is the defined 
>>>> bootjdk) works on all the other platforms.
>>>>
>>>> /Erik
>>>>
>>>
>>
>