Re: Memory leak in ios.cc:326

[Carlo Wood <carlo at alinoe dot com>]

On Fri, Mar 28, 2003 at 09:29:42AM -0600, Benjamin Kosnik wrote:
> >I have my hands full at the moment, but I wanted to ACK this message.
> 
> Me too. I'm going to take a break from the 27_io testsuite and scope
> this a bit today.
> 
> There's one test case for callbacks, I'm using that. Looking at this on
> 3.2, with GLIBCPP_FORCE_NEW, everything looks ok. But on 3.4, I'm seeing
> some issues with constructing the locale cache for the standard streams.
> I suspect this is also the issue for 3.3.

I can investigate this further as soon as I have libcwd
working for 3.3.  But that can take weeks, because there
is a problem with the fact that a basic_stringstream
with a custom allocator STILL calls std::allocator, not
for the storage for the stringstream, but for for example
the locale object, an auto_ptr and an allocation that
can happen in operator<<(ostream&, int).

I'll have to redesign things a bit because so far I relied
on an object with a custom allocator NOT to allocate memory
in any other way than by using _that_ allocator :/

-- 
Carlo Wood <carlo at alinoe dot com>