https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124551
Bug ID: 124551
Summary: Exception-safety issue in
pmr::synchronized_pool_resource due to bad_alloc
during thread exit
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libstdc++
Assignee: unassigned at gcc dot gnu.org
Reporter: redi at gcc dot gnu.org
Target Milestone: ---
The thread-local pools managed by a std::pmr::synchronized_pool_resource are
destroyed on thread exit by:
// Called when a thread exits
extern "C" {
static void destroy_TPools(void* p)
{
using _TPools = synchronized_pool_resource::_TPools;
_TPools::destroy(static_cast<_TPools*>(p));
}
}
which calls:
static void destroy(_TPools* p)
{
exclusive_lock l(p->owner._M_mx);
// __glibcxx_assert(p != p->owner._M_tpools);
p->move_nonempty_chunks();
Which does:
// move all non-empty chunks to the shared _TPools
for (int i = 0; i < owner._M_impl._M_npools; ++i)
for (auto& c : pools[i]._M_chunks)
if (!c.empty())
shared[i]._M_chunks.insert(std::move(c), r);
The insert call might need to grow the capacity of the _Pool::vector which
might throw bad_alloc. That would (probably?) terminate the process, because
we're running thread-specific destructor functions during thread exit.
