clang and clang++ with ASAN generate different output - c++

Question

I'm trying to add ASAN (Google's/Clang's address sanitize) to our project and stuck at this problem.
For example, we have this simple C++ code
#include <iostream>
int main() {
std::cout << "Started Program\n";
int* i = new int();
*i = 42;
std::cout << "Expected i: " << *i << std::endl;
}
Then, I build it with clang++
clang++-3.8 -o memory-leak++ memory_leak.cpp -fsanitize=address -fno-omit-frame-pointer -g
The program gives this output
Started Program
Expected i: 42
=================================================================
==14891==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 4 byte(s) in 1 object(s) allocated from:
#0 0x4f2040 in operator new(unsigned long) (memory-leak+++0x4f2040)
#1 0x4f4f00 in main memory_leak.cpp:4:11
#2 0x7fae13ce6f44 in __libc_start_main /build/eglibc-SvCtMH/eglibc-2.19/csu/libc-start.c:287
SUMMARY: AddressSanitizer: 4 byte(s) leaked in 1 allocation(s).
Cool, it works and symbolizer gives meaningful information too.
Now, I build this with clang
clang-3.8 -o memory-leak memory_leak.cpp -std=c++11 -fsanitize=address -fno-omit-frame-pointer -g -lstdc++
And the program gives this output
Started Program
Expected i: 42
=================================================================
==14922==ERROR: LeakSanitizer: detected memory leaks
Direct leak of 4 byte(s) in 1 object(s) allocated from:
#0 0x4c3bc8 in malloc (memory-leak+0x4c3bc8)
#1 0x7f024a8e4dac in operator new(unsigned long) (/usr/lib/x86_64-linux-gnu/libstdc++.so.6+0x5edac)
#2 0x7f0249998f44 in __libc_start_main /build/eglibc-SvCtMH/eglibc-2.19/csu/libc-start.c:287
SUMMARY: AddressSanitizer: 4 byte(s) leaked in 1 allocation(s).
Ok, it detects memory leak, but the stack trace looks strange and it doesn't really include the memory_leak.cpp:4:11 line.
I've spent quite a while trying to narrow down this problem in our codebase and eventually, the only difference, is clang vs clang++.
Why it's event a problem, can't we use clang++?
We use bazel, which uses CC compiler instead of CXX for some blah-balh reasons. We cannot blindly force to use it CXX because we have CC dependencies which cannot be build by CXX. So...
Any idea how to get the same ASAN output when used with clang and clang++? Or, how to make Bazel to use clang++ for C++ targets and clang for C targets?

Answer 1

This seems to be a bug in Clang, could you file bug report in their tracker? (EDIT: this was [resolved as not-a-bug](Asan developers https://github.com/google/sanitizers/issues/872) so probly needs to be fixed by Bazel developers instead).
Some details: when you use ordinary clang, it decides not to link C++ part of Asan runtime as can be seen in Tools.cpp:
if (SanArgs.linkCXXRuntimes())
StaticRuntimes.push_back("asan_cxx");
and SanitizerArgs.cpp:
LinkCXXRuntimes =
Args.hasArg(options::OPT_fsanitize_link_cxx_runtime) || D.CCCIsCXX();
(note the D.CCCIsCXX part, it checks for clang vs. clang++ whereas instead they need to check the file type).
C++ part of the runtime contains interceptor for operator new so this would explain why it's missing when you link with clang instead of clang++. On a positive side, you should be able to work around this by adding -fsanitize-link-c++-runtime to your flags.
As for the borked stack, by default Asan unwinds stack with frame pointer based unwinder which has problems unwinding through code which wasn't built with -fno-omit-frame-pointer (like libstdc++.so in your case). See e.g. this answer for another example of such behavior and available workarounds.