Re: [Qemu-devel] [PATCH 2/3] glib: add g_thread_new() compat function

[Michael Tokarev <mjt@tls.msk.ru>]

02.05.2014 15:08, Peter Maydell wrote:
> On 2 May 2014 12:01, Daniel P. Berrange <berrange@redhat.com> wrote:
>> On Fri, May 02, 2014 at 02:52:23PM +0400, Michael Tokarev wrote:
>>> Stefan Hajnoczi:
>>>> +#if !GLIB_CHECK_VERSION(2, 31, 0)
>>>> +static inline GThread *g_thread_new(const gchar *unused,
>>>> +                                    GThreadFunc func,
>>>> +                                    gpointer data)
>>>> +{
>>>> +    GThread *thread = g_thread_create(func, data, TRUE, NULL);
>>>> +    if (!thread) {
>>>> +        g_error("g_thread_create failed");
>>>> +    }
>>>> +    return thread;
>>>> +}
>>>> +#endif
>>>
>>> About g_error():
>>>
>>> "This function will result in a core dump; don't use it for errors you expect.
>>>  Using this function indicates a bug in your program, i.e. an assertion failure."
>>>
>>> I'm not sure if this is like an assertion failure, probably yes.  But we should
>>> not, I think, dump core in this situation.  Is there a glib function that does
>>> (fatal) error reporting but does NOT dump core?
>>
>> I think that's what  g_critical is intended for. It is more severe than
>> g_warning, but won't exit or abort the process.
> 
> I'm not convinced we should be emitting any kind of
> warning message here anyway -- surely it's up to the
> caller to handle thread creation failure? The glib
> warning/error functions presumably print to stderr,
> which has all the usual issues with possibly messing
> up guest output.

Actually the whole point is moot.  Here's what g_thread_new() does (in 2.31+):

GThread *
g_thread_new (const gchar *name,
              GThreadFunc  func,
              gpointer     data)
{
  GError *error = NULL;
  GThread *thread;

  thread = g_thread_new_internal (name, g_thread_proxy, func, data, 0, &error);

  if G_UNLIKELY (thread == NULL)
    g_error ("creating thread '%s': %s", name ? name : "", error->message);

  return thread;
}

So that's what we should use in our g_thread_new() impl,
and this is what Stefan used, in a slightly less complex way.

I'll add this to my wrapper too.

Thanks,

/mjt