NonSFI NaCl: Plumb Exception IRT enough for breakpad.

chrome/test/data/nacl/crash/nonsfi_crash_via_signal_handler.cc

+// Copyright 2014 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+

[Mark Seaborn, 2014/04/23 03:49:50]

Could you name this file after the interface it tests, i.e. irt_exception,
please?  (The "signal handler" part is a detail of how irt_exception is
implemented.)

Something like chrome/test/data/nacl/irt_exception_test.cc.

Could you enable this for SFI NaCl too, since it's not specific to Non-SFI NaCl?

[Junichi Uekawa, 2014/04/23 07:56:15]

Done.
Done.

gyp looks ugly, I think we probably need to fix generate_nmf.

+#include <sched.h>
+#include <unistd.h>
+
+#include "native_client/src/include/nacl/nacl_exception.h"
+#include "native_client/src/shared/platform/nacl_check.h"

[Mark Seaborn, 2014/04/23 03:49:50]

I don't think you use this?

[Junichi Uekawa, 2014/04/23 07:56:15]

Done.

+#include "ppapi/native_client/tests/ppapi_test_lib/test_interface.h"
+
+namespace {
+
+static void MyNaClExceptionHandler(struct NaClExceptionContext *context) {
+  printf("--- MyNaClExceptionHandler\n");
+  TEST_PASSED;
+  // Wait for the javascript to realize that we have succeeded. If we
+  // exit from here we shall crash and it's hard to distinguish that

[Mark Seaborn, 2014/04/23 03:49:50]

You could exit using setjmp()/longjmp() to exit from this fault handler.  That
might be a little cleaner than doing an infinite loop.

See exception_test.c in the NaCl tree, which does something similar.

[Junichi Uekawa, 2014/04/23 07:56:15]

Done.

+  // crash from a complete failure to register a fault handler.
+  while (1) {
+    sched_yield();
+  }
+}
+
+void CrashViaSignalHandler() {
+  printf("--- CrashViaSignalHandler\n");
+
+  int retval = nacl_exception_set_handler(MyNaClExceptionHandler);
+  if (retval != 0) {
+    printf("Unexpected return value from nacl_exception_set_handler: %d\n",
+           retval);
+    TEST_FAILED;
+    return;
+  }
+  CRASH;

[Mark Seaborn, 2014/04/23 03:49:50]

I think it would be better to use __builtin_trap() -- that's a bit more direct
than reusing CRASH.

[Junichi Uekawa, 2014/04/23 07:56:15]

It seemed to me that CRASH is more consistent with other tests in
//chrome/test/data/..., but either way is fine with me.

+}
+
+}  // namespace
+
+void SetupTests() {
+  RegisterTest("CrashViaSignalHandler", CrashViaSignalHandler);
+}
+
+void SetupPluginInterfaces() {
+

[Mark Seaborn, 2014/04/23 03:49:50]

Nit: remove empty line at start of function

[Junichi Uekawa, 2014/04/23 07:56:15]

Done.

+  // none

[Mark Seaborn, 2014/04/23 03:49:50]

Nit: maybe remove this?  Empty functions aren't usually commented.

[Junichi Uekawa, 2014/04/23 07:56:15]

Done.

+}