[gwp_asan] Employ EXPECT_DEATH for zxtest compatibility

[Caslyn]

Summary

Employ a similar tactic introduced by https://reviews.llvm.org/D94362
for gwp_asan tests. zxtest ASSERT_DEATH syntax differs from gtest in
that it expects a lambda.

zxtest does not have EXPECT_DEATH, so it introduced for Fuchsia builds
and wraps the expression with a lambda to create a compatible syntax
between zxtest and gtest for death tests.

An example of where this compatiblity is needed is in
never_allocated.cpp.

[hctim]

compiler-rt/lib/gwp_asan/tests/harness.h

20

Y?

21–25

are there scenarios when ASSERT_DEATH is not defined? seems odd.

[Caslyn]

compiler-rt/lib/gwp_asan/tests/harness.h

20

The second arg to zxtest's ASSERT_DEATH differs from gtest in that it is only used as an (optional) message arg for debugging if the crash did not happen. Since it doesn't do any regex matching on the exception, Y loses its meaning in this way, and an empty string is passed. We could omit the second arg to this call altogether - though the reason it's added is to have some parity with the implementation for scudo tests (https://github.com/llvm/llvm-project/blob/main/compiler-rt/lib/scudo/standalone/tests/scudo_unit_test.h#L23).

21–25

That's a good question - I can only think if zxtest removes it from its API. @mcgrathr, does this existence check for zxtest's ASSERT_DEATH in scudo/gwp_asan do anything for us?

[hctim]

compiler-rt/lib/gwp_asan/tests/harness.h

20

ah, yeah i forgot that fuchsia does the sane thing with crashes (registering an external death-handling process) rather than signal handlers, where you can check for an expected bit of text.

[Caslyn]

compiler-rt/lib/gwp_asan/tests/harness.h

21–25

gentle ping @mcgrathr

[Caslyn]

Removed unnecessary ASSERT_DEATH existance check

[Caslyn]

compiler-rt/lib/gwp_asan/tests/harness.h

21–25

In the interest of moving this CL forward to unblock Fuchsia's gwp-asan rolls, removed the existence check.

[hctim]

LGTM w/ nit.

compiler-rt/lib/gwp_asan/tests/harness.h

18

// zxtest defines a different ASSERT_DEATH, taking a lambda and an error message
// if death didn't occur, versus gtest taking a statement and a string to search
// for in the dying process. zxtest doesn't define an EXPECT_DEATH, so we use
// that in the tests below (which works as intended for gtest), and we define
// EXPECT_DEATH as a wrapper for zxtest's ASSERT_DEATH. Note that zxtest drops
// the functionality for checking for a particular message in death.

Clarifies:

1. What the args are in zxtest (which aren't clear to us not familiar with zxtest),
2. How we can abuse gtest defining EXPECT_DEATH but we can wrap it under zxtest, and
3. We ignore the error message on zxtest.

[Caslyn]

Amend comment

compiler-rt/lib/gwp_asan/tests/harness.h

 #define GWP_ASAN_TESTS_HARNESS_H_
 
 #include <stdarg.h>
 
 #if defined(__Fuchsia__)
 #define ZXTEST_USE_STREAMABLE_MACROS
 #include <zxtest/zxtest.h>
 namespace testing = zxtest;
+// zxtest defines a different ASSERT_DEATH, taking a lambda and an error message
+// if death didn't occur, versus gtest taking a statement and a string to search
+// for in the dying process. zxtest doesn't define an EXPECT_DEATH, so we use
+// that in the tests below (which works as intended for gtest), and we define
+// EXPECT_DEATH as a wrapper for zxtest's ASSERT_DEATH. Note that zxtest drops
+// the functionality for checking for a particular message in death.
+#define EXPECT_DEATH(X, Y) ASSERT_DEATH(([&] { X; }), "")
 #else
 #include "gtest/gtest.h"
 #endif
 
 #include "gwp_asan/guarded_pool_allocator.h"
 #include "gwp_asan/optional/backtrace.h"
 #include "gwp_asan/optional/printf.h"
 #include "gwp_asan/optional/segv_handler.h"
 #include "gwp_asan/options.h"

compiler-rt/lib/gwp_asan/tests/never_allocated.cpp

TEST_P(BacktraceGuardedPoolAllocatorDeathTest, NeverAllocated) {
   std::string DeathNeedle =
       "GWP-ASan cannot provide any more information about this error";
 
   // Trigger a guard page in a completely different slot that's never allocated.
   // Previously, there was a bug that this would result in nullptr-dereference
   // in the posix crash handler.
   char *volatile NeverAllocatedPtr = static_cast<char *>(Ptr) + 0x3000;
   if (!Recoverable) {
-    ASSERT_DEATH(*NeverAllocatedPtr = 0, DeathNeedle);
+    EXPECT_DEATH(*NeverAllocatedPtr = 0, DeathNeedle);
     return;
   }
 
   *NeverAllocatedPtr = 0;
   CheckOnlyOneGwpAsanCrash(GetOutputBuffer());
   ASSERT_NE(std::string::npos, GetOutputBuffer().find(DeathNeedle));
 
   // Check that subsequent invalid touches of the pool don't print a report.