Re: New warnings in libstdc++-v3 since 2004-01-05

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Andreas Jaeger <aj@suse.de> writes:

| I see the following new warnings when compiling GCC CVS on different
| Linux platforms:
| ../../../../libstdc++-v3/src/bitmap_allocator.cc: In member function 'size_t* __gnu_cxx::free_list::_M_
| get(size_t)':
| ../../../../libstdc++-v3/src/bitmap_allocator.cc:110: warning: control reaches end of non-void function
| 
| ../../../../libstdc++-v3/testsuite/testsuite_abi.cc: In function ‘symbol& get_symbol(const std::string&, const symbols&)’:
| ../../../../libstdc++-v3/testsuite/testsuite_abi.cc:268: warning: control reaches end of non-void function
| 
| What is the best way to fix this?

fix the compiler regression.

| 
| Should we add to include/bits/functexcept.h some
| __attribute__((noreturn)) ?

No!!!  The function does return!

This is a compiler regression.  The body of the function is:

     _M_get(size_t __sz) throw(std::bad_alloc)
     {
   #if defined __GTHREADS
       _Lock __bfl_lock(&_S_bfl_mutex);
       __bfl_lock._M_lock();
   #endif
       iterator __temp =
         __gnu_cxx::balloc::__lower_bound
         (_S_free_list.begin(), _S_free_list.end(),
          __sz, _LT_pointer_compare());

       if (__temp == _S_free_list.end() || !_M_should_i_give(**__temp,
       __sz))
         {
           // We release the lock here, because operator new is
           // guaranteed to be thread-safe by the underlying
           // implementation.
   #if defined __GTHREADS
           __bfl_lock._M_unlock();
   #endif
           // Try twice to get the memory: once directly, and the 2nd
           // time after clearing the free list. If both fail, then
           // throw std::bad_alloc().
           int __ctr = 2;
           while (__ctr)
             {
               size_t* __ret = 0;
               --__ctr;
               try
                 {
                   __ret = reinterpret_cast<size_t*>
                     (::operator new(__sz + sizeof(size_t)));
                 }
               catch(...)
                 {
                   this->_M_clear();
                 }
               if (!__ret)
                 continue;
               *__ret = __sz;
               return __ret + 1;
             }
           std::__throw_bad_alloc();
         }
       else
         {
           size_t* __ret = *__temp;
           _S_free_list.erase(__temp);
   #if defined __GTHREADS
           __bfl_lock._M_unlock();
   #endif
           return __ret + 1;
         }
     }

In the if-part, the function either (a) returns, from the while-loop;
or (b) throws.  In the else-part, it does return.

Please, fill a C++ front-end PR.

-- Gaby
PS: I would not mind removing the exception-specification from 
the function.