[lit] Support the `%shared_libasan` lit substitution on Apple platforms.

[delcypher]

Summary

The previous value looks Linux specific so that has been guarded with
the host OS being Linux.

On Apple platforms %shared_libasan expands to the absolute path of the
ASan dylib.

Previously on Linux %shared_libasan expanded to just the file name
of the shared library rather than the absolute path to the library.
This is likely a bug because it would rely on the OS's dynamic linker
to find the shared library which could accidentally pick up a system copy
rather than the shared library that was just built.

For other platforms we emit a warning if config.asan_dynamic is true.

This patch also only defines the substitution when config.asan_dynamic
is true because using this substitution only makes sense when the
dynamic library is available.

[kubamracek]

What breaks if we switch to the absolute path on Linux?

test/asan/lit.cfg

108–113

This is unnecessarily expanded. Fold at least the .format() call into one line (personally I'd have this whole thing on one line).

[phosek]

I'd be fine with using an absolute path on all platforms, it's what the Clang driver does as well so this certainly looks like a bug.

[delcypher]

In D53111#1261287, @kubamracek wrote:

What breaks if we switch to the absolute path on Linux?

I'm not sure. I will test and report back.

test/asan/lit.cfg

108–113

Sure. I'll fix this.

[delcypher]

In D53111#1262521, @delcypher wrote:

In D53111#1261287, @kubamracek wrote:

What breaks if we switch to the absolute path on Linux?

I'm not sure. I will test and report back.

@kubamracek Linux tests seem to pass if I use an absolute path so I'll make the change for Linux too.

[delcypher]

• Try to fix formatting
• Use absolute path on Linux too
• Emit a warning for other platforms when config.asan_dynamic is true.
• Only set substitution when config.asan_dynamic is true.

[phosek]

LGTM

[delcypher]

In D53111#1262958, @phosek wrote:

LGTM

@phosek Thanks for the review. I'll wait for a LGTM from @kubamracek or @george.karpenkov so they can check the Apple side of this.

[eugenis]

I think this might break android which runs tests at a different absolute path.

[delcypher]

In D53111#1263563, @eugenis wrote:

I think this might break android which runs tests at a different absolute path.

Hmm now that I think about it, that's very likely. I don't have an Android device I can test with so I can't really investigate this. How about I add an extra condition that checks for config.android and if that's true, use the old behaviour (filename rather than an absolute path) along with a FIXME comment?

[eugenis]

nvm, all tests using this substitution are disabled on Android anyway

test/asan/lit.cfg

 
 def build_invocation(compile_flags):
   return " " + " ".join([config.clang] + compile_flags) + " "
 
 config.substitutions.append( ("%clang ", build_invocation(target_cflags)) )
 config.substitutions.append( ("%clangxx ", build_invocation(target_cxxflags)) )
 config.substitutions.append( ("%clang_asan ", build_invocation(clang_asan_cflags)) )
 config.substitutions.append( ("%clangxx_asan ", build_invocation(clang_asan_cxxflags)) )
+if config.host_os == 'Linux':
+  # FIXME: This should be an absolute path!
   config.substitutions.append( ("%shared_libasan", "libclang_rt.asan%s.so" % config.target_suffix))
+elif config.host_os == 'Darwin':
+  shared_libasan_path = os.path.join(
+    config.compiler_rt_libdir,
+    'libclang_rt.asan_{}_dynamic.dylib'.format(
+      config.apple_platform
+    )
+  )
+  config.substitutions.append( ("%shared_libasan", shared_libasan_path) )
+
 if config.asan_dynamic:
   config.substitutions.append( ("%clang_asan_static ", build_invocation(clang_asan_static_cflags)) )
   config.substitutions.append( ("%clangxx_asan_static ", build_invocation(clang_asan_static_cxxflags)) )
 
 # Windows-specific tests might also use the clang-cl.exe driver.
 if platform.system() == 'Windows':
   clang_cl_cxxflags = ["-Wno-deprecated-declarations",
                        "-WX",