In article ,
Roger Davis wrote:
>I am trying to use nothrow new() and am encountering what seems
>to be a bug in the SGI/Irix C++ environment. The following

This is the subject of bug-id 862202.
(We added your report to that bug.)

Latest EDG front ends give a warning message about this
source code of yours, but we are not yet sure if this
warning is real or simply a mistaken (on our part)
integration of EDG front-end source.

I'm not a C++ language lawyer, but if I had to guess I'd guess
this is likely our bug, either in our runtime or our integration
of the EDG front end.

Sorry.
David Anderson [Not an official SGI spokesperson...]
PS: google 'nothrow new' to find the gotw (Guru Of
The Week) notes on nothrow. Before you use it too much :-)

PPS: IRIX CC has -LANG:exceptions=OFF to turn C++
into an older-style (before exceptions) so new returns 0
on failure. In IRIX virtual swap (like in Linux)
can mean malloc will succeed apparently but reference to
the memory can fail (Irrelevant Historical note: IRIX created
virtual swap before Linux had the concept).

In article ,
Roger Davis wrote:
> I am trying to use nothrow new() and am encountering what seems
> to be a bug in the SGI/Irix C++ environment. The following
> program runs OK under Solaris and Linux, but dumps core at the
> delete[] statement under Irix. Is this really an SGI implementation bug,
> or have I done something stupid? (Other than trying to use nothrow
> new(), that is ;-) )
>
> #include
> #include
>
> std::nothrow_t NoThrow;
>
> class Junk {
> public:
> Junk();
> ~Junk();
> };
>
> Junk::Junk() {
> (void) fprintf(stderr, "constr\n"); (void) fflush(stderr);
> }
>
> Junk::~Junk() {
> (void) fprintf(stderr, "destr\n"); (void) fflush(stderr);
> }
>
> main() {
> Junk *jnk;
>
> if ((jnk= new(NoThrow) Junk[5]) == (Junk *) 0) {
> (void) fprintf(stderr, "memory allocation error\n");
> exit(-1);
> }
>
> delete[] jnk;
> }
>
> My debugger stack looks like this:
>
> _kill()
> _raise()
> abort()
> _array_pointer_not_from_vec_new()
> __array_delete_general2()
> __vec_delete2()
> main() <---- stopped on the delete[] statement
>
> If I use regular new() instead, i.e.,
>
> jnk= new Junk[5];
>
> then the delete[] works OK under Irix. Do I need to be using some
> special form of delete[] to indicate that the new() was called with
> nothrow? I have seen no documentation anywhere that indicates that
> I should, but maybe I am not looking in the right places.
>
> I also tried
>
> float *f;
>
> f= new(NoThrow) float[100];
> delete[] f;
>
> under Irix and that works OK. The problem seems to occur only when
> using delete[] on a non-primitive.

This may be an issue with unsupported placement array new on
an older compiler.

Please provide the compiler/linker versions, and the c++_eoe
and compiler_eoe versions, and the compile/link command.

And the IRIX OS version (just because).

Joe
--
------------------------------------------------------------------------
Joseph Michaud - SGI Apps Eng 978-562-8894 x483-8894 jmichaud@sgi.com

> Please provide the compiler/linker versions, and the c++_eoe
> and compiler_eoe versions, and the compile/link command.

% CC -version
MIPSpro Compilers: Version 7.3.1.3m

% ld -V
ld32: INFO 153: Version 7.30.

c++_eoe 7.3.1.3m
compiler_eoe 7.3.1.3m

% CC -o junk -g junk.cc
% ./junk
constr
constr
constr
constr
constr
Abort (core dumped)

> And the IRIX OS version (just because).

eoe IRIX Execution Environment, 6.5.16m

(Thanks for looking into this, Joe.)