Enable lsan test suite for darwin x86-64

[alekseyshl]

test/lsan/lit.common.cfg

79

I think it would be more sustainable to come up with the readable way to merge both Linux (line 71) and Darwin conditions into one.

[fjricci]

I've been running this on some large internal FB programs, and I've found a couple issues with dyld memory which I think stem from the fact that sanitizer_procmaps_mac behaves differently than sanitizer_procmaps_linux. I'm also not very pleased with the performance, although I haven't dug into that too much yet.

I think I'll modify this patch to still enable the tests (to prevent regressions), but leave lsan disabled by default until I'm more confident things are working well.

[fjricci]

Disable by default, but keep testing enabled

[alekseyshl]

test/lsan/lit.common.cfg

73

Too many negatives :) maybe break it up?

supported_linux = config.host_os is 'Linux' and config.host_arch in ['x86_64', 'mips64', 'arm', 'armhf', 'armv7l']
supported_darwin = config.host_os is 'Darwin' and config.target_arch is 'x86_64'
if not (supported_linux or supported_darwin):
  config.unsupported = True

[fjricci]

test/lsan/lit.common.cfg

73

Didn't even think to use intermediate variables - will do

[fjricci]

Refactor to address comments

[kubamracek]

test/asan/lit.cfg

44–45

Does the ASan test suite pass with that? Specifically, the tests in Testcases/Darwin/?

[fjricci]

test/asan/lit.cfg

44–45

They do on my local macbook, yes.

44–45

(The failures there were addressed by this patch a while back: D31978)

[fjricci]

Looks like this caused a large number of test failures. I'm a bit confused, since things worked fine locally, but I'm going to try to do a totally clean build with the buildbot config to see if I can repro.

[fjricci]

The failures still don't repro with a clean build, but it looks like I can just download the buildbot state from here: http://green.lab.llvm.org/green/job/clang-stage1-configure-RA/ws/, so hopefully I can cherry-pick into that and see the failures.

[fjricci]

Well, looks like the zips that get downloaded are invalid zips, and extraction fails midway through. I've been looking into the actual failures, and it's actually a bit interesting. It's almost as if the tests are being run on a pretty old version of lsan, as pretty much all of the failures look like things that I've seen before and fixed (ie false-positive leaks from libxpc, etc).

[fjricci]

(this is the test run, by the way: http://green.lab.llvm.org/green/job/clang-stage1-configure-RA_check/30443)

[fjricci]

Oh wow, it is. If you look in the source-code browser, it appears to be running on out-of-date code (note the CHECK(0 && "unimplemented") in StopTheWorld): http://green.lab.llvm.org/green/job/clang-stage1-configure-RA/ws/compiler-rt.src/lib/lsan/lsan_common_mac.cc

[fjricci]

nvm, looks like the source browser is now up-to-date

[kubamracek]

Can you take a look athttp://lab.llvm.org:8080/green/job/clang-stage1-configure-RA_check/31273/console?

[fjricci]

Yeah, I xfail-ed those two tests for now to fix the buildbots, since I figure it's worth the test coverage (and this doesn't enable lsan by default outside the test suite anyway). Looking into the failures currently.

[fjricci]

I'm running the tests in a "repeat while test passes" loop, and neither of them are failing for me on 10.11 or 10.12.

[fjricci]

Looks like XFAIL doesn't work, because they pass in LeakSanitizer-AddressSanitizer, and are just very flaky in LeakSanitizer-Standalone, so they end up XPASS-ing. I'll move them to UNSUPPORTED to fix the buildbot, and then we can go from there. I don't think it's reverting all of the lsan test suite for 2 flaky tests that I can't repro on 10.11 or 10.12.

cmake/config-ix.cmake

 if (COMPILER_RT_HAS_SANITIZER_COMMON AND DFSAN_SUPPORTED_ARCH AND
     OS_NAME MATCHES "Linux")
   set(COMPILER_RT_HAS_DFSAN TRUE)
 else()
   set(COMPILER_RT_HAS_DFSAN FALSE)
 endif()
 
 if (COMPILER_RT_HAS_SANITIZER_COMMON AND LSAN_SUPPORTED_ARCH AND
-    OS_NAME MATCHES "Linux|FreeBSD")
+    OS_NAME MATCHES "Darwin|Linux|FreeBSD")
   set(COMPILER_RT_HAS_LSAN TRUE)
 else()
   set(COMPILER_RT_HAS_LSAN FALSE)
 endif()
 
 if (COMPILER_RT_HAS_SANITIZER_COMMON AND MSAN_SUPPORTED_ARCH AND
     OS_NAME MATCHES "Linux")
   set(COMPILER_RT_HAS_MSAN TRUE)

test/asan/lit.cfg

 # Platform-specific default ASAN_OPTIONS for lit tests.
 default_asan_opts = ''
 if config.host_os == 'Darwin':
   # On Darwin, we default to `abort_on_error=1`, which would make tests run
   # much slower. Let's override this and run lit tests with 'abort_on_error=0'.
   # Also, make sure we do not overwhelm the syslog while testing.
   default_asan_opts = 'abort_on_error=0'
   default_asan_opts += ':log_to_syslog=0'
+
+  # On Darwin, leak checking is not enabled by default. Enable for x86_64
+  # tests to prevent regressions
+  if config.target_arch == 'x86_64':
+    default_asan_opts += ':detect_leaks=1'
 elif config.android:
   # The same as on Darwin, we default to "abort_on_error=1" which slows down
   # testing. Also, all existing tests are using "not" instead of "not --crash"
   # which does not work for abort()-terminated programs.
   default_asan_opts = 'abort_on_error=0'
 
 if default_asan_opts:
   config.environment['ASAN_OPTIONS'] = default_asan_opts

 if config.target_arch != 'arm' and config.target_arch != 'armhf' and config.target_arch != 'aarch64':
   config.available_features.add('stable-runtime')
 
 # Fast unwinder doesn't work with Thumb
 if re.search('mthumb', config.target_cflags) is not None:
   config.available_features.add('fast-unwinder-works')
 
 # Turn on leak detection on 64-bit Linux.
-if config.host_os == 'Linux' and (config.target_arch == 'x86_64' or config.target_arch == 'i386'):
+if ((config.host_os == 'Linux') and (config.target_arch == 'x86_64' or config.target_arch == 'i386')) \
+  or ((config.host_os == 'Darwin') and (config.target_arch == 'x86_64')):
   config.available_features.add('leak-detection')
 
 # Set LD_LIBRARY_PATH to pick dynamic runtime up properly.
 push_dynamic_library_lookup_path(config, config.compiler_rt_libdir)
 
 # GCC-ASan uses dynamic runtime by default.
 if config.compiler_id == 'GNU':
   gcc_dir = os.path.dirname(config.clang)

test/lsan/lit.common.cfg

 def build_invocation(compile_flags):
   return " " + " ".join([config.clang] + compile_flags) + " "
 
 config.substitutions.append( ("%clang ", build_invocation(clang_cflags)) )
 config.substitutions.append( ("%clangxx ", build_invocation(clang_cxxflags)) )
 config.substitutions.append( ("%clang_lsan ", build_invocation(clang_lsan_cflags)) )
 config.substitutions.append( ("%clangxx_lsan ", build_invocation(clang_lsan_cxxflags)) )
 
-# LeakSanitizer tests are currently supported on x86-64 Linux, arm Linux and mips64 Linux only.
-if config.host_os not in ['Linux'] or config.host_arch not in ['x86_64', 'mips64', 'arm', 'armhf', 'armv7l']:
+# LeakSanitizer tests are currently supported on x86-64 Linux, arm Linux, mips64 Linux, and x86_64 Darwin
+if (config.host_os not in ['Linux'] or config.host_arch not in ['x86_64', 'mips64', 'arm', 'armhf', 'armv7l'])
+   and (config.host_os is not 'Darwin' or config.target_arch is not 'x86_64'):
   config.unsupported = True
 
 # Don't support Thumb due to broken fast unwinder
 if re.search('mthumb', config.target_cflags) is not None:
   config.unsupported = True
 
 config.suffixes = ['.c', '.cc', '.cpp']