UBSan

UBSan, short for Undefined Behavior Sanitizer, is a runtime analysis tool in the LLVM project used with the Clang compiler to detect undefined behavior in C, C++, and other languages supported by LLVM during program execution. It works by instrumenting generated code to insert checks that validate assumptions required by the language standard; when a check fails, UBSan emits a diagnostic with the location in the source and the type of undefined behavior, and may terminate the program depending on configuration.

UBSan focuses on runtime detection of undefined behavior rather than memory errors. Common checks include signed integer overflow, integer division by zero, shifts that are out of range or overflow, null pointer dereference, misaligned memory access, and out-of-bounds array indexing. The set of detectors can vary by language version and toolchain, and newer releases may add additional checks for other UB scenarios. It is designed to complement other sanitizers in the LLVM ecosystem, such as AddressSanitizer and ThreadSanitizer, by targeting semantic violations rather than memory safety alone.

Usage typically involves compiling with -fsanitize=undefined and running the program as usual to obtain diagnostics. Recovery behavior can be controlled with flags like -fsanitize-recover=undefined to continue after a first error, or by setting the UBSan runtime option via the UBSAN_OPTIONS environment variable (for example to enable stack traces). Instrumentation adds runtime overhead, so UBSan is commonly used in development and testing rather than production builds.

Overall, UBSan provides a practical means to surface subtle undefined behaviors in C and C++ code, aiding debugging and verification efforts without requiring static guarantees.