On Wed, 8 Dec 2021 at 00:36, Jonathan Wakely wrote:
>
> On Tue, 7 Dec 2021 at 21:52, Florian Weimer wrote:
> >
> > * Jonathan Wakely:
> >
> > > On Tue, 7 Dec 2021, 21:20 Florian Weimer via Libstdc++, <libstdc++@gcc.gnu.org>
> > > wrote:
> > >
> > >  * Jonathan Wakely via Libstdc:
> > >
> > >  > If necessary we could keep the terminate-on-cancellation behaviour as
> > >  > _ZNSt18condition_variable4waitERSt11unique_lockISt5mutexE@GLIBCXX_3.4.11
> > >  > and export the new behaviour as @@GLIBCXX_3.4.30, although this patch
> > >  > doesn't do that.
> > >
> > >  Note that if this fix escapes into the wild and then you have to make
> > >  the symbol version change, you will break newer applications.  In a few
> > >  cases in glibc, we proactively added aliases at a different symbol
> > >  version, but with the same implementation (at first).
> > >
> > > To be safe, we probably should preserve the old behaviour for the old
> > > version of the symbol. If we decide that the new behaviour is always
> > > preferable, we could change that later by making the old symbol an
> > > alias for the new. If we don't decide that, we'll be glad we made it a
> > > separate symbol.
> >
> > On the other hand, with separate versions, it's possible to reintroduce
> > the old behavior at a later date, as a bugfix.  It's not strictly
> > necessary to do that work upfront.  It's just nice to have this option.
>
> Ah yes, a new symbol version gives us more flexibility in every direction.
>
> > > I'll see if I can get it working with two versioned symbols. We don't
> > > actually do that in libstdc++ currently, we only have a single version
> > > of every symbol.
> >
> > Ping me if you want to discuss options. 8->
>
> Thanks. I'll try it and let you know how I get on.

After resolving a PEBKAC issue, here's an incremental diff that
preserves the old behaviour for the existing @GLIBCXX_3.4.11 symbol,
but adds a new @@GLIBCXX_3.4.30 symbol that supports cancellation via
__forced_unwind.

Maybe we should also do this in the implementation of the old noexcept function:

__attribute__((used))
void
__nothrow_wait_cv::wait(std::unique_lock<std::mutex>& lock) noexcept
{
  int old;
  int err = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
  this->condition_variable::wait(lock);
  if (!err && old != PTHREAD_CANCEL_DISABLE)
    pthread_setcancelstate(old, &old);
}

This would prevent cancellation from terminating a process if it uses
the old symbol. So we'd have a new symbol that supports cancellation,
and an old one that safely disables it.