[lldb/test][Darwin] Ask dyld where the real python is

[vsk]

Summary

On macOS, we can't do the DYLD_INSERT_LIBRARIES trick with a shim
python binary as the ASan interceptors get loaded too late. Find the
"real" python binary, copy it, and invoke it.

Hopefully this makes the GreenDragon and swift-ci sanitizer bots
happy...

I tested this out by running `../llvm-macosx-x86_64/bin/llvm-lit test
--filter TestNSDictionarySynthetic.py` in an ASanified swift-lldb build
directory and it worked (i.e. no more "interceptors loaded too late"
messages).

[JDevlieghere]

Thanks Vedant for coming up with a structural solution to this problem!

lldb/test/API/lldbtest.py

37

This would have to run under the Python in executable, not the one lit is running with.

[vsk]

Query dyld while running within the right python process.

[JDevlieghere]

lldb/test/API/lldbtest.py

86

SIP still matters here, which is why we need to copy the python binary and not just invoke it with the result from get_darwin_real_python.py. Can you merge the two comments?

I also think that we might set DYLD_INSERT_LIBRARIES unconditionally on other hosts. Can we guard this whole thing with a check for macOS?

[vsk]

• Reinstated comment about SIP to explain why we copy python
• Added a platform check

[JDevlieghere]

LGTM!

lldb/test/API/get_darwin_real_python.py

+# On macOS, system python binaries like /usr/bin/python and $(xcrun -f python3)
+# are shims. They do some light validation work and then spawn the "real" python
+# binary. Find the "real" python by asking dyld -- sys.executable reports the
+# wrong thing more often than not. This is also useful when we're running under
+# a Homebrew python3 binary, which also appears to be some kind of shim.
+def getDarwinRealPythonExecutable():
+    import ctypes
+    dyld = ctypes.cdll.LoadLibrary('/usr/lib/system/libdyld.dylib')
+    namelen = ctypes.c_ulong(1024)
+    name = ctypes.create_string_buffer(b'\000', namelen.value)
+    dyld._NSGetExecutablePath(ctypes.byref(name), ctypes.byref(namelen))
+    return name.value.decode('utf-8').strip()
+
+print(getDarwinRealPythonExecutable())

lldb/test/API/lldbtest.py

 from __future__ import absolute_import
 import os
 import tempfile
 import subprocess
 import sys
+import platform
 
 import lit.Test
 import lit.TestRunner
 import lit.util
 from lit.formats.base import TestFormat
 
 def getBuildDir(cmd):
     found = False
def mkdir_p(path):
     if not os.path.isdir(path):
         raise OSError(errno.ENOTDIR, "%s is not a directory"%path)
 
 class LLDBTest(TestFormat):
     def __init__(self, dotest_cmd):
         self.dotest_cmd = dotest_cmd
 
     def getTestsInDirectory(self, testSuite, path_in_suite, litConfig,
                             localConfig):
         source_path = testSuite.getSourcePath(path_in_suite)
         for filename in os.listdir(source_path):
             # Ignore dot files and excluded tests.
             if (filename.startswith('.') or filename in localConfig.excludes):
                 continue
 
             # Ignore files that don't start with 'Test'.
             if not filename.startswith('Test'):
def execute(self, test, litConfig):
         # On Windows, the system does not always correctly interpret
         # shebang lines.  To make sure we can execute the tests, add
         # python exe as the first parameter of the command.
         cmd = [executable] + self.dotest_cmd + [testPath, '-p', testFile]
 
         builddir = getBuildDir(cmd)
         mkdir_p(builddir)
 
-        # The macOS system integrity protection (SIP) doesn't allow injecting
-        # libraries into system binaries, but this can be worked around by
-        # copying the binary into a different location.
+        # On macOS, we can't do the DYLD_INSERT_LIBRARIES trick with a shim
+        # python binary as the ASan interceptors get loaded too late. Also,
+        # when SIP is enabled, we can't inject libraries into system binaries
+        # at all, so we need a copy of the "real" python to work with.
+        #
+        # Find the "real" python binary, copy it, and invoke it.
         if 'DYLD_INSERT_LIBRARIES' in test.config.environment and \
-                (executable.startswith('/System/') or \
-                executable.startswith('/usr/bin/')):
+                platform.system() == 'Darwin':
             copied_python = os.path.join(builddir, 'copied-system-python')
             if not os.path.isfile(copied_python):
                 import shutil, subprocess
                 python = subprocess.check_output([
                     executable,
-                    '-c',
-                    'import sys; print(sys.executable)'
+                    os.path.join(os.path.dirname(os.path.realpath(__file__)),
+                        'get_darwin_real_python.py')
                 ]).decode('utf-8').strip()
                 shutil.copy(python, copied_python)
             cmd[0] = copied_python
 
         if 'lldb-repro-capture' in test.config.available_features or \
            'lldb-repro-replay' in test.config.available_features:
             reproducer_root = os.path.join(builddir, 'reproducers')
             mkdir_p(reproducer_root)