Comment 1:

You may have to runtime.LockOSThread to interact with GTK's event loop?

Comment 2:

1. This code doesn't use GTK's event loop. It's explicitly started with gtk_main usually.
2. Other code works fine (20+ demos using different widgets and events).
3. I tried runtime.LockOSThread(). And tried running the code above in the 'init'
function. Doesn't help.

Comment 3:

I believe that gtk is creating some thread and then that thread
gets a signal, and then the Go signal handler is invoked.
I am not sure what to do about this.  We like our signal handlers,
but they can't cope with being invoked on non-Go threads.
We could ignore such signals easily enough, but perhaps
gtk is really trying to handle that signal (or maybe it's a SIGSEGV
or something).
Russ

Labels changed: added priority-go1, removed priority-triage.

Owner changed to builder@golang.org.

Status changed to Accepted.

Comment 4:

Just want to mention, that the issue is most likely related to gtk DBus usage. On the
client side, it seems that the only signal it touches is SIGPIPE. It has code:
  #if HAVE_DECL_MSG_NOSIGNAL
  static dbus_bool_t _dbus_modify_sigpipe = FALSE;
  #else
  static dbus_bool_t _dbus_modify_sigpipe = TRUE;
  #endif
And then on connection opening it does:
  if (_dbus_modify_sigpipe)
    _dbus_disable_sigpipe ();
Which in turn results in a function call (if true):
  void _dbus_disable_sigpipe (void)
  {
    signal (SIGPIPE, SIG_IGN);
  }
On my machine I know it doesn't run _dbus_disable_sigpipe, maybe that's the issue.
Honestly I'm not an expert on how signals work in linux.

Comment 5:

Can you run your program under strace -f to find which signal is being
delivered?
If it is only SIGPIPE, we might be able to do a simple workaround for Go 1.

Comment 6:

The worst part that it runs fine under strace/ltrace. But it died once, however only
once, I wasn't able to repeat that under strace, see the second segfault log file. But
I'm afraid it won't be very helpful.

Attachments:

1. strace-log.txt (160911 bytes)
2. strace-log-segfault.txt (153031 bytes)

Comment 7:

Hm.. it dies often if I run strace without "-o" option (writes output to a file), here's
the two variants of dying:
SIGPIPE:
[pid 11326] read(3,
"\1\10\v\0\22\0\0\0\37\0\0\0\0\0\0\0H\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 4096) = 104
[pid 11326] read(3, 0x89d5b08, 4096)    = -1 EAGAIN (Resource temporarily unavailable)
[pid 11326] read(3, 0x89d5b08, 4096)    = -1 EAGAIN (Resource temporarily unavailable)
[pid 11326] poll([{fd=3, events=POLLIN|POLLOUT}], 1, -1) = 1 ([{fd=3, revents=POLLOUT}])
[pid 11326] writev(3, [{"\24\0\6\0\1\0@\1\212\1\0\0\6\0\0\0\0\0\0\0\4\0\0\0", 24},
{NULL, 0}, {"", 0}], 3) = 24
[pid 11326] poll([{fd=3, events=POLLIN}], 1, -1) = 1 ([{fd=3, revents=POLLIN}])
[pid 11326] read(3, "\1
\f\0\1\0\0\0\6\0\0\0\0\0\0\0\1\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 4096) = 36
[pid 11326] read(3, 0x89d5b08, 4096)    = -1 EAGAIN (Resource temporarily unavailable)
[pid 11326] read(3, 0x89d5b08, 4096)    = -1 EAGAIN (Resource temporarily unavailable)
[pid 11326] write(1, "unix:abstract=/tmp/dbus-LOsdVe90"..., 73) = -1 EPIPE (Broken pipe)
[pid 11326] --- {si_signo=SIGPIPE, si_code=SI_USER, si_pid=11326, si_uid=1000,
si_value={int=3076563288, ptr=0xb760a158}} (Broken pipe) ---
Process 11315 resumed
Process 11326 detached
Process 11315 detached
SIGCHLD? (but here it seems almost finished, we can see the last close calls):
[pid 11678] write(1, "unix:abstract=/tmp/dbus-LOsdVe90"..., 73) = 73
[pid 11678] write(1, "\370)\0\0", 4)    = 4
[pid 11678] write(1, "\1\0@\1", 4)      = 4
[pid 11678] close(1)                    = 0
[pid 11678] close(2)                    = 0
[pid 11678] exit_group(0)               = ?
Process 11678 detached
[pid 11677] <... select resumed> )      = 1 (in [9])
[pid 11677] --- {si_signo=SIGCHLD, si_code=CLD_EXITED, si_pid=11678, si_status=0,
si_utime=0, si_stime=0} (Child exited) ---
[pid 11677] --- {si_signo=SIGSEGV, si_code=SEGV_MAPERR, si_addr=0x2c} (Segmentation
fault) ---
Process 11677 detached
[pid 11676] +++ killed by SIGSEGV +++
[pid 11675] +++ killed by SIGSEGV +++
[pid 11674] +++ killed by SIGSEGV +++
+++ killed by SIGSEGV +++
And normally it runs fine:
read(3, 0x9842b08, 4096)                = -1 EAGAIN (Resource temporarily unavailable)
read(3, 0x9842b08, 4096)                = -1 EAGAIN (Resource temporarily unavailable)
poll([{fd=3, events=POLLIN|POLLOUT}], 1, -1) = 1 ([{fd=3, revents=POLLOUT}])
writev(3, [{"\24\0\6\0\1\0@\1\212\1\0\0\6\0\0\0\0\0\0\0\4\0\0\0", 24}, {NULL, 0}, {"",
0}], 3) = 24
poll([{fd=3, events=POLLIN}], 1, -1)    = 1 ([{fd=3, revents=POLLIN}])
read(3, "\1 \f\0\1\0\0\0\6\0\0\0\0\0\0\0\1\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 4096) = 36
read(3, 0x9842b08, 4096)                = -1 EAGAIN (Resource temporarily unavailable)
read(3, 0x9842b08, 4096)                = -1 EAGAIN (Resource temporarily unavailable)
write(1, "unix:abstract=/tmp/dbus-LOsdVe90"..., 73) = 73
write(1, "\370)\0\0", 4)                = 4
write(1, "\1\0@\1", 4)                  = 4
close(1)                                = 0
close(2)                                = 0
exit_group(0)                           = ?
Process 11746 detached

Comment 8:

I created CL 5797068 to at least diagnose the problem better.
I do not believe we will be able to fix this for Go 1.

Labels changed: added priority-later, removed priority-go1.

Comment 9:

[nsf @ go-test]$ ./test
runtime: signal received on thread not created by Go.
Segmentation fault
Clearly that's the case.. Will look forward to a fix.

Comment 10 by joshrickmar:

I've run into this issue as well, trying to add a GtkEntry to a container (code
attached) with GTK 3.8 on OpenBSD.  I'm running go tip (changeset 8519983c00e8), and the
process no longer crashes, but produces a more useful error message:
runtime: signal received on thread not created by Go: SIGCHLD: child status has changed
Is there any way with the current source that we can catch this and ignore it, since we
no longer crash?

Attachments:

1. test.go (297 bytes)

Comment 11:

as a workaround, you could just add a return statement in function runtime.badsignal in
src/pkg/runtime/os_$GOOS.c to ignore any signals received on foreign threads.

Comment 12 by joshrickmar:

With the exception of processes ignoring my SIGINTs, that's a pretty good fix.
Here's the signals that I'm seeing with GTK and my test, now that it's not quitting
immediately:
runtime: signal received on thread not created by Go: SIGTERM: termination
runtime: signal received on thread not created by Go: SIGWINCH: window size change
Both of these signals, as well as a few others, have the default action of being ignored
(according to signal(3)).  Should this be fixed by listening for these signals, and if
they are sent, to ignore them completely?

Comment 13 by joshrickmar:

Oops, that should be:
runtime: signal received on thread not created by Go: SIGCHLD: child status has changed
runtime: signal received on thread not created by Go: SIGWINCH: window size change

Comment 14 by joshrickmar:

Here's a quick patch I put together, that ignores those signals that have no default
action.  I only modified the OpenBSD files, but the other platforms should have a
similar fix.  I've had no issues making GTK3 calls with cgo with this patch.

Attachments:

1. ignore-signals.patch (1609 bytes)

Comment 15:

This issue was closed by revision 2f1ead7.

Status changed to Fixed.

CL https://golang.org/cl/12503 mentions this issue.