Re: Why a try/catch in priority_queue::push/pop?!?

[Howard Hinnant <hhinnant at apple dot com>]

On Aug 3, 2006, at 2:42 PM, Paolo Carlini wrote:

> Hi,
>
> today I noticed by chance these strange things: for example, push  
> (likewise pop):
>
>      void
>      push(const value_type& __x)
>      {
>    try
>        {
>          c.push_back(__x);
>          std::push_heap(c.begin(), c.end(), comp);
>        }
>    catch(...)
>        {
>          c.clear();
>          __throw_exception_again;
>        }
>      }
>
> why that try/catch? Certainly I cannot find support in the Standard  
> for its presence and, AFAICS, can also run against reasonable user  
> expectations: i.e., if c.push_back throws (e.g., bad_alloc) the  
> user may reasonably expect that c remains unchanged! The case of  
> push_heap is different, more complex, but again, I don't think the  
> user, basing on the letter of the Standard can generally (i.e.,  
> portably) expect that a clear() will be invoked (supposedly to do  
> him the "favor" of a "consistent", read, empty! , container)
>
> I think the try/catch is very old, coming from the HP/SGI STL,  
> maybe I missing historical facts which I'd like to know... Anyone  
> helping?

<shakes head> Can't blame this one on HP/SGI, just checked, no-try/ 
catch there. :-)

I agree with your analysis of the code and the standard.  I think the  
user could reasonably expect priority_queue<int>::push to have the  
strong exception guarantee, though I note that the standard doesn't  
quite say that (that I can find).

-Howard