[v3] PR 53270 fix hppa-linux bootstrap regression

[Jonathan Wakely <jwakely dot gcc at gmail dot com>]

We've known for ages that it's not portable to do:

__gthread_mutex_t tmp = __GTHREAD_MUTEX_INIT;
_M_mutex = __tmp;

As PR 53270 shows, the copy assignment now actually fails in C++11
mode on platforms using LinuxThreads, because the mutex has a volatile
member so in C++11 mode the copy assignment operator is deleted.

This patch changes <ext/concurrence.h> to use a
brace-or-equals-initializer for the mutexes and condition variable in
C++11 mode, allowing hppa-linux to bootstrap again and avoiding the
non-portable construct everywhere (this is already the approach we
take for std::mutex etc. in <mutex>). It also makes the same change to
the mutex used in <ext/rope> and fixes a resource leak in that header
by ensuring the mutex is destroyed when it was initialized by
__gthread_mutex_init_function.

        PR libstdc++/53270
        * include/ext/concurrence.h (__mutex, __recursive_mutex, __cond): Use
        NSDMI in C++11 mode.
        * include/ext/rope (_Refcount_Base): Likewise. Destroy mutex in
        destructor when initialized by function.

Tested x86_64-linux, powerpc64-linux, hppa-linux and x86_64-netbsd (on
the 4.7 branch instead, as netbsd doesn't bootstrap at the moment.)
Committed to trunk.

For 4.6.4 and 4.7.2 I plan to make a less intrusive change, #undef'ing
the __GTHREAD_MUTEX_INIT, _GTHREAD_RECURSIVE_MUTEX_INIT and
__GTHREAD_COND_INIT macros on hppa-linux in C++11 mode, so that the
init functions are used instead.  This fixes the bootstrap regression
on hppa-linux without affecting other targets.

Attachment: patch.txt
Description: Text document