(11) RFR (S) JDK-8196880: VS2017 Addition of Global Delete Operator with Size Parameter Conflicts with Arena's Chunk Provided One

[Lois Foltan]

On 2/14/2018 9:26 AM, Thomas Stüfe wrote:

> Hi Lois,
>
> thanks for fixing this!
Thanks for the review!

>
> Small nit, not part of your patch, but still:
>
> Chunks get allocated via CHeapObj::operator new() but deleted in 
> Chunk::chop() with raw ::free(). Would it not be cleaner to call 
> CHeapObj::operator delete() instead (it does a free() too, but that 
> would be symmetrical? That would require that we actually implement 
> Chunk::operator delete(), I guess.
Sure, that seems reasonable and logical.  However, note it still doesn't 
alleviate the need to define a delete operator that takes a size 
parameter.  At the point where placement new is used, the compiler tries 
to find an appropriate matching operator delete and would error (C4291) 
with a failure to find one if not provided.

Thanks,
Lois

>
> Best Regards, Thomas
>
>
> On Wed, Feb 14, 2018 at 2:48 PM, Lois Foltan <lois.foltan at oracle.com 
> <mailto:lois.foltan at oracle.com>> wrote:
>
>     Please review this change in VS2017 to the delete operator due to
>     C++14 standard conformance.  From
>     https://msdn.microsoft.com/en-us/library/mt723604.aspx
>     <https://msdn.microsoft.com/en-us/library/mt723604.aspx>
>
>     The function|void operator delete(void *, size_t)|was a placement
>     delete operator corresponding to the placement new function "void
>     * operator new(size_t, size_t)" in C++11. With C++14 sized
>     deallocation, this delete function is now a/usual deallocation
>     function/(global delete operator). The standard requires that if
>     the use of a placement new looks up a corresponding delete
>     function and finds a usual deallocation function, the program is
>     ill-formed.
>
>     Thank you to Kim Barrett for proposing the fix below.
>
>     open webrev at
>     http://cr.openjdk.java.net/~lfoltan/bug_jdk8196880/webrev/
>     <http://cr.openjdk.java.net/%7Elfoltan/bug_jdk8196880/webrev/>
>     bug link https://bugs.openjdk.java.net/browse/JDK-8196880
>     <https://bugs.openjdk.java.net/browse/JDK-8196880>
>
>     Testing complete (hs-tier1-3, jdk-tier1-3)
>
>     Thanks,
>     Lois
>
>
>