Re: Wanted: if new returns NULL, don't run constructor

[mrs at wrs dot com (Mike Stump)]

> Date: Fri, 26 Feb 1999 13:05:43 -0700
> From: Greg McGary <gkm@eng.ascend.com>
> To: egcs@egcs.cygnus.com
> Cc: balsup@eng.ascend.com, gkm@eng.ascend.com

> In a legacy embedded system that's tight on memory, I can't afford
> exception handling.

-fno-exceptions -fno-rtti will get you what you want.

> Neither is it useful to handle memory allocation failures with a
> new_handler.  What I'd like to see is `new OBJ' return NULL on
> failure,

Please buy a new book on C++.  In C++ you can say new (nothrow) OBJ
and have it do what you want.

For legacy code you can

#define new new (nothrow)

and get mostly what you want.

> and for OBJ's constructor to detect (this==NULL) and return
> immediately.

If all the above fails, you can use -fcheck-new.  RTFM:

@item -fcheck-new
Check that the pointer returned by @code{operator new} is non-null
before attempting to modify the storage allocated.  The current Working
Paper requires that @code{operator new} never return a null pointer, so
this check is normally unnecessary.

An alternative to using this option is to specify that your
@code{operator new} does not throw any exceptions; if you declare it
@samp{throw()}, g++ will check the return value.  See also @samp{new
(nothrow)}.