See where you use LibC the most.
Trace calls failing tests. Then - roast!

LibSee is a single-file library for profiling LibC calls and 🔜 fuzzy testing. To download and compile the script and run your favorite query:

gcc -g -O2 -fno-builtin -fPIC -nostdlib -nostartfiles -shared -o libsee.so libsee.c

LibSee overrides LibC symbols using LD_PRELOAD, profiling the most commonly used functions, and, optionally, fuzzing their behavior for testing. The library yields a few binaries when compiled:

libsee.so # Profiles LibC calls
libsee_and_knee.so # Correct LibC behavior, but fuzzed!

Tricks Used

There are several things worth knowing, that came handy implementing this.

• One way to implement this library would be to override the _start symbols, but implementing correct loading sequence for a binary is tricky, so I use conventional dlsym to lookup the symbols on first function invocation.
• On x86_64 architecture, the rdtscp instruction yields both the CPU cycle and also the unique identifier of the core. Very handy if you are profiling a multi-threaded application.
• Once the unloading sequence reaches libsee.so, the STDOUT is already closed. So if you want to print to the console, you may want to reopen the /dev/tty device before printing usage stats.
• Calling convention for system calls on Aarch64 and x86 differs significantly. On Aarch64 I use the generalized openat with opcode 56. On x86 it's opcode 2.
• On MacOS the sprintf, vsprintf, snprintf, vsnprintf are macros. You have to #undef them.
• On Release builds compilers love replacing your code with memset and memcpy calls. As the symbol can't be found from inside LibSee, it will SEGFAULT so don't forget to disable such optimizations for built-ins -fno-builtin.
• No symbol versioning is implemented, vanilla dlsym is used over the dlvsym.

Coverage

LibC standard is surprisingly long, so not all of the functions are covered. Feel free to suggest PRs covering the rest:

• memory management
• byte-strings
• algorithms
• date and time
• input/output
• wide-character strings
• concurrency and atomics
• retrieving error numbers
• numerics
• multi-byte strings
• wide-character IO
• localization
• anything newer than C 11

There are a few other C libraries that most of the world reuses, rather than implementing from scratch in other languages:

• BLAS and LAPACK
• PCRE RegEx
• hsearch, tsearch, and pattern matching extensions

Program support utilities aren't intended.