On Oct 29, 2012, at 9:32 AM, Jonathan Wakely wrote:
> On Oct 29, 2012 1:54 PM, "Perry Smith" wrote:
>>
>> My original logic in adding them to libstdc++.a is they are only used
>> (called) by code automatically produced by g++ and not gcc.
>
> That doesn't make sense. Both gcc and g++ are just driver programs
> that invoke the appropriate compiler program, which for C++ is
> cc1plus. So C++ code can be compiled by gcc, and doesn't necessarily
> link to libstdc++.
>
> Putting the files in libsupc++ might make sense but libstdc++ is
> definitely wrong.
You've mentioned this before. I think you are needlessly splitting hairs. The
construct that produces the code that causes the failure is a c++ construct.
The construct that causes the compiler to produce a call to atexit is a static
of a variable with a non-trivial destructor declared inside a function. This
program, does not compile with gcc:
> #include <stdio.h>
>
> class foo {
> private:
> const char *name;
> void mess(const char *s) {
> printf("%s %s\n", s, name);
> }
>
> public:
> foo(const char *s) {
> name = s;
> mess("ctor");
> }
>
> ~foo() {
> mess("dtor");
> }
> };
>
> foo& func()
> {
> static foo val("func");
> return val;
> }
But it compiles just fine with g++.
For most platforms, the compiler produces a call to __cxa_atexit. On AIX,
because AIX does not have __cxa_atexit natively, the compiler produces a call
to atexit. atexit doesn't really have the smarts required for shared
libraries. I've discussed this all at length before.
Perry
