🚀 go-pugleaf

In article <1992Oct28.013753.12311@nuscc.nus.sg> scstech@solomon.technet.sg (Fong Khai Yin) writes:
>I think there is a bug in g++ 2.2.2 where an object is used as an argument to a function
>that is expecting a char*.
>Please correct me if this is incorrect.

No, it's not a bug.  Temporary objects may be deleted after their last
use; however, cfront (and Sun C++) doesn't delete them until the end
of the enclosing scope.  

g++ destroys temporaries fairly aggressively, which is legal according
to the ARM but surprising when used with objects with (const char*)
casts defined.

>Is it possible to get a patch for this problem ???
No, because you would then port your code to yet another compiler
and have the same problem.  Your program is not portable.

The only way according to the current rules to guarantee that a temporary
lives until the end of a block is to give the temporary a name.

Example: a typical String class, that defines an operator const char*
function.

Given the declarations

extern "C" FILE * fopen(const char*,const char*);
String foo();

the code

	FILE* fd = fopen(foo(),"r");

is unsafe, because the temporary object returned by foo() may be
destroyed before fopen is called.  The only portable way to write
the above is something like

	String& tmp = foo();	// give the temporary a name
	FILE* fd = fopen(tmp,"r");

The ANSI C++ standardization committee is still debating the rules
for destruction of temporaries.  According to the present rules,
both the cfront and the g++ behavior is legal.

--
Joe Buck	jbuck@ohm.berkeley.edu