Re: Critical ~__pool troubles (libstdc++/17937)

[Matt Austern <austern at apple dot com>]

On Oct 11, 2004, at 10:37 AM, Paolo Carlini wrote:

> Hi everyone,
>
> earlier today (exactly while I was looking more into Carlo message)
> Michael sent me this testcase, distilled from KDE arts:
>
> #include <string>
> #include <list>
> using namespace std;
> static list<string> modulePath;
> int main()
> {
>  modulePath.push_back("hallo");
>  return 0;
> }
>
> Therefore, no special settings are needed (e.g., per_type_pools) to
> show that something is still not ok with ~__pool. I'm attaching below
> valgrind's output for a --enable-__cxa_atexit build (removing it
> avoids, as expected, all the errors).
>
> Frankly, I'm wondering whether we really, *really*, /really/ need to
> explicitly deallocate the pools instead of demanding the work to the
> OS: why the original SGI/HP allocator didn't do that?? Perhaps Matt
> could help here??

There's not very much to say.  (But if you want *really* expert opinion
you should ask Hans Boehm, not me.  This is an area where the HP
and SGI implementations were completely different; Hans rewrote
memory allocation from scratch.)

It's pretty much what you already said: nobody could think of a good
reason to deallocate the pool explicitly, since it would  only happen
at the end of the program and at that point the OS would reclaim all
the storage anyway.  All you'll be doing is dirtying pages and making
program exit slower, so why bother?

			--Matt