Yet another XLC bug: wrong handling of array new operator with default arguments

[Volker Simonis]

Yes, Goetz has temporarily switched off event logging (
http://hg.openjdk.java.net/ppc-aix-port/jdk8/hotspot/rev/3dc08c0a6867) and
we hope this construct will not be used soon in other parts of the VM.

Nevertheless we would really appreciate a fix:)

On Fri, Jun 7, 2013 at 4:16 PM, Steve Poole <spoole at linux.vnet.ibm.com>wrote:

> Oh dear - ok, I'll pass it along.   I assume you've got a work around for
> this at the moment?
>
>
> On 6 Jun 2013, at 17:47, Volker Simonis <volker.simonis at gmail.com> wrote:
>
> Hi Steve,
>
> during the merge with the latest HotSpot version Goetz found yet another
> bug in XLC.
>
> I managed to break it down into the following minimal C++ program
> ('array_new.cpp' - see attachment).
>
> In the example, the size of a Foo<Buffer> obejct should be 42 bytes and
> consequently the size of an array of 10 Foo<Buffer> obejcts should always
> be 420 bytes. However in the first allocation of the example program, the
> array new allocator of Foo gets called with a size argument of 10 (which is
> the arity of the array to allocate) instead of its size (which would be
> 420). This is wrong, because it allocates to few storage for the array.
> This leads to crashes if the array will be accessed later on.
>
> $ xlc++ array_new.cpp
> $ ./a.out
> sizeof(Foo<Buffer>[10]) = 420
> *Foo::new [](size=10, foo=0)*
> Foo::new [](size=420, foo=123)
> sizeof(Foo<Buf>[10]) = 420
> Foo::new [](size=420, foo=0)
> Foo::new [](size=420, foo=123)
>
> Notice that all the other allocations are successful because they either
> explicitly specify the default argument of the array new operator or
> because  they use POD data type (struct 'Buf' which has no constructor
> instead of struct 'Buffer') as template argument for Foo.
>
> The error occurs with both XLC 12:
>
> IBM XL C/C++ for AIX, V12.1 (5765-J02, 5725-C72)
> Version: 12.01.0000.0003
>
> and XLC 10:
>
> IBM XL C/C++ for AIX, V10.1
> Version: 10.01.0000.0015
>
> Could you please open a bug report for this issue and urge your XLC
> colleagues to fix this issue as fast as possible because this is a
> show-stopper for our current plan to merge our PPC/AIX HotSpot port into
> the the master repository.
>
> Thank you and best regards,
> Volker
>
> PS: in HotSpot this issue appears in http://hg.openjdk.java.net/hsx/hotspot-main/hotspot/file/tip/src/share/vm/utilities/events.hpp
> where the event loggers are initialized:
>
>        88   EventLogBase<T>(const char* name, int length =
> LogEventsBufferEntries):
>        89     _name(name),
>        90     _length(length),
>        91     _count(0),
>        92     _index(0),
>        93     _mutex(Mutex::event, name) {
>  *      94     _records = new EventRecord<T>[length];*
>        95   }
>
> because 'EventLogBase' is derived from 'CHeapObj' which has an array new
> operator with an additional default argument. It finally leads to crashes
> in various places where the loggers are used during runtime.
>
> <array_new.cpp>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/ppc-aix-port-dev/attachments/20130607/c0e95649/attachment.html