RFR: 8247396: PCH should be disabled for opt-size feature with clang after JDK-8246751

[Erik Joelsson]

On 2020-06-11 06:41, Magnus Ihse Bursie wrote:
> On 2020-06-11 14:34, Erik Joelsson wrote:
>> I agree with the assessment, but I think the disabling should happen 
>> in configure. That way the configure log can accurately tell the user 
>> why it was disabled. This can be done by just adding another elif in 
>> BPERF_SETUP_PRECOMPILED_HEADERS. To check for a jvm feature in 
>> configure, you can use JVM_FEATURES_IS_ACTIVE(opt-size).
> Actually, I'm not sure I agree with the assessment. Why does this only 
> happen for clang? Is it a problem when building with opt-size and pch 
> for other compilers?
>
> If not, that's likely due to the problematic files being excluded from 
> PCH (if I remember correctly). I believe a better way forward would be 
> to make sure that files that are compiled with a different 
> optimization level should be excluded from PCH for clang as well.
>
Right, I just took a look and in JvmOverrideFiles.gmk we build a list of 
files to exclude from precompiled header. We could just add 
$(OPT_SPEED_SRC) to this list (as that variable is defined earlier in 
JvmFeatures.gmk). OPT_SPEED_SRC is empty unless opt-size is true, so 
just adding it to the list should be fine.

/Erik

> /Magnus
>>
>> /Erik
>>
>> On 2020-06-11 02:25, jiefu(傅杰) wrote:
>>> Hi all,
>>>
>>> JBS:    https://bugs.openjdk.java.net/browse/JDK-8247396
>>> Webrev: http://cr.openjdk.java.net/~jiefu/8247396/webrev.00/
>>>
>>> -O3 is used for clang on MacOS after JDK-8246751.
>>> But it breaks the builds of VMs with opt-size feature when PCH is 
>>> enabled.
>>>
>>> The reason is that the PCH was built with -Os while some other files 
>>> were built with -O3, which causes an error with clang.
>>> It would be better to disable the PCH for opt-size builds.
>>>
>>> Thanks a lot.
>>> Best regards,
>>> Jie
>