AddressSanitizer (Clang Sanitizers)

Edit page on GitHub

The AddressSanitizer is integrated into Clang 3.4+ and therefore, if Clang is installed, using it with Mono’s autogen.sh is as simple as running the following:

$ cd /home/root/of/mono
$ ./autogen.sh --prefix=/usr/local CC=clang CFLAGS=-fsanitize=address LDFLAGS=-fsanitize=address CXX=clang
$ make
...

Please keep in mind that CXX has to be set to clang to avoid unwanted side effects and compiling errors.

After autogen.sh has been executed as suggested above, the following make command uses Clang and injects necessary reporting functions.

In the context of Mono, the AddressSanitizer (without the LeakSanitizer, see below) produces a lot less false positives than the ThreadSanitizer, which makes working with it more straight forward.

Dynamic Blacklisting

Still, false positives need to be hidden, and so far, the best option is to work with blacklists. Please have a look at this article to find out how to work with blacklists in combination with Mono.

Since there are only a handful of false positives, no macros or special functions can be used to permanently blacklist anything directly within the code (August 2017).

Runtime Options

Runtime options can be provided by using the environment variable ASAN_OPTIONS. Please find detailed information about runtime options and Mono in this article and all available options on Google’s GitHub page. The following two options prove to be particularly useful:

Ignore the Leaks

By default, the AddressSanitizer starts the LeakSanitizer as soon as the program terminates otherwise successfully. In the context of Mono this means several 100 (sometimes up to a few 1000) reported leaks. detect_leaks=0 tells the AddressSanitizer to shut down without starting the LeakSanitizer.

Never Stop

If not told otherwise, the AddressSanitizer exits the program after detecting the first error. While this is helpful when debugging a specific issue, it can be an unwanted side effect when trying to find all problems of a program at once. Using halt_on_error=false can help there.