JDK 17: First Release Candidate

[David Holmes]

Hi Florian,

Just to follow up on a different point ...

On 14/08/2021 7:23 pm, Florian Weimer wrote:
> * David Holmes:
> 
>> Hi Matthias,
>>
>> On 14/08/2021 2:00 am, Matthias Klose wrote:
>>> On 8/6/21 7:09 PM, mark.reinhold at oracle.com wrote:
>>>> There are no unresolved P1 bugs in build 35, so that is our
>>>> first JDK 17 Release Candidate.
>>>>
>>>> Binaries available here, as usual: https://jdk.java.net/17
>>> this fails to build with glibc 2.34, i.e. on recent Fedora and
>>> Ubuntu
>>> development versions:
>>> /<<PKGBUILDDIR>>/test/hotspot/jtreg/runtime/StackGuardPages/exeinvoke.c:
>>> In
>>> function 'set_signal_handler':
>>> /<<PKGBUILDDIR>>/test/hotspot/jtreg/runtime/StackGuardPages/exeinvoke.c:72:15:
>>> error: storage size of 'altstack' isn't constant
>>>      72 |   static char altstack[SIGSTKSZ];
>>>         |               ^~~~~~~~
>>> gmake[4]: *** [test/JtregNativeHotspot.gmk:1525:
>>> /<<PKGBUILDDIR>>/build/support/test/hotspot/jtreg/native/support/exeinvoke/exeinvoke.o]
>>> Error 1
>>
>> That seems like a glibc bug to me as SIGSTKSZ is supposed to be a constant.
> 
> It is still a constant in glibc's POSIX mode, but OpenJDK does not use
> POSIX mode (with good reason).

Can you elaborate on what that good reason is please? If this is 
something critical then it really needs to be documented somewhere. It 
makes me uncomfortable that we have spent a lot of time cleaning up 
hotspot runtime code to provide shared "posix" implementations, if we 
may not even be getting Posix semantics.

Thanks,
David

>  8192 bytes is really tight on AVX-512
> capability CPUs, so the code could still be buggy if it compiled.
> 
> Thanks,
> Florian
>