Implement Dump-on-DCHECK (via a new LogSeverity).

base/test/gtest_util.h

Index: base/test/gtest_util.h
diff --git a/base/test/gtest_util.h b/base/test/gtest_util.h
index 8dfb1f236f38713e3cf5985c7974cd3f16429b3f..c8e44e7aaa10a390eaa2643c2d87903aad8cb22e 100644
--- a/base/test/gtest_util.h
+++ b/base/test/gtest_util.h
@@ -21,7 +21,7 @@
 // happen and as such executing the statement results in undefined behavior
 // (|statement| is compiled in unsupported configurations nonetheless).
 // Death tests misbehave on Android.
-#if DCHECK_IS_ON() && defined(GTEST_HAS_DEATH_TEST) && !defined(OS_ANDROID)
+#if DCHECK_IS_ON() && defined(GTEST_HAS_DEATH_TEST) && !defined(OS_ANDROID) && !defined(DCHECK_IS_DUMP_WITHOUT_CRASH)

[Wez, 2016/12/19 23:57:46]

This causes EXPECT_DCHECK_DEATH to be a no-op if DCHECK is defined to
dump-on-DCHECK. For some tests we could still have them work, and just verify
that dump-without-crashing gets called.  For others the DCHECKs actually protect
against e.g. C++ data-structure corruption, so we'll crash anyway if we don't
have DCHECK crash.

Q: Is there any down-side to losing this test coverage in dump-on-DCHECK builds?

 
 // EXPECT/ASSERT_DCHECK_DEATH tests verify that a DCHECK is hit ("Check failed"
 // is part of the error message), but intentionally do not expose the gtest
@@ -42,7 +42,7 @@
     GTEST_AMBIGUOUS_ELSE_BLOCKER_ \
     if (::testing::internal::AlwaysTrue()) { \
       GTEST_LOG_(WARNING) \
-          << "Death tests are not supported on this platform.\n" \
+          << "Death tests are not supported in this configuration.\n" \
           << "Statement '" #statement "' cannot be verified."; \
     } else if (::testing::internal::AlwaysFalse()) { \
       ::testing::internal::RE::PartialMatch(".*", (regex)); \