Re: [commit/windows] Add thread ID in SuspendThread error warning message.

[Christopher Faylor <cgf-use-the-mailinglist-please@sourceware.org>]

On Tue, Jun 18, 2013 at 04:43:49PM -0700, Joel Brobecker wrote:
>> FWIW, I've seen this from time to time and have convinced myself that
>> it's a red herring.  When it happens, it probably should just silently
>> continue.
>
>That's interesting. I have the following patch in AdaCore's tree
>which I have been uhming and ahming about. Would it apply to your
>situation as well?
>
>-                   warning (_("SuspendThread failed. (winerr %u)"),
>-                            (unsigned) err);
>-                   return NULL;
>+                   /* If SuspendThread failed with error 5 (access
>+                      denied), then ignore the error.  It's unclear
>+                      where this comes from and how to prevent it.
>+                      But in the meantime, ignoring it seems to allow
>+                      us to inspect the thread (including fetching
>+                      registers) without apparent ill effect.  */
>+                   if (err != 5)
>+                     {
>+                       warning (_("SuspendThread (tid=0x%x) failed."
>+                                  " (winerr %d)"),
>+                                (unsigned) id, (unsigned) err);
>+                       return NULL;
>+                     }

That's basically what I'm doing in the Cygwin release:

                if (SuspendThread (th->h) == (DWORD) -1)
                  {
                    DWORD err = GetLastError ();
-                   warning (_("SuspendThread failed. (winerr %u)"),
-                            (unsigned) err);
+                   /* Can get a ERROR_INVALID_HANDLE if the main thread has
+                      exited. */
+                   if (err != ERROR_INVALID_HANDLE)
+                     warning (_("SuspendThread(%p) failed. (winerr %u)"),
+                              (void *) th->h, (unsigned) err);
                    return NULL;
                  }
                th->suspended = 1;


>I think there are other situations as well were we emit a warning,
>and where I've been considering the idea of downgrading them to
>complaints, so that users don't unnecessarily get concerned. But
>I wanted to investigate a little bit before doing so, and I've
>never had the time :-(.

I think this probably shouldn't be a warning.  If there is an
issue then it will become clearer at some other point.

cgf