NumPy Debug Build and Testing Setup

This guide explains how to clone NumPy, build it with debug symbols, run tests, and debug the C code using gdb.

1. Setup

Run the following bash file to build the debug environment.

./setup.sh

You can also use spin to build the environment as mentioned in the official website. I prefer the vanilla approach.

./setup.sh --spin

If the arguments are provided correctly, the script will automatically:

• create the required Python virtual environment,
• install all necessary dependencies,
• configure the development environment,
• and build NumPy with the appropriate debug configuration.

After the script finishes successfully, the environment will be ready for building, running tests, and debugging.

2. Debugging with GDB

Start Virtual Enviroment and GDB with Python

source numpy-debug312/bin/activate
gdb python3.12

Stop when shared libraries are loaded

For example, if you are debugging the legacy_random_binomial_original you might need to look for a shared library _generator.cpython-312-x86_64-linux-gnu.so by continuing.

(gdb) set stop-on-solib-events 1
(gdb) run check_rnd.py
(gdb) continue

Alternatively, set breakpoints directly in NumPy C code

(gdb) set breakpoint pending on
(gdb) break legacy_random_binomial_original
(gdb) run check_rnd.py
(gdb) continue

---

Notes

• NumPy must be compiled with debug flags (-O0 -g) for effective debugging (look inside setup.sh)
• Breakpoints can be set directly in NumPy C functions once the shared libraries are loaded.
• When probing into the random module keep it in mind that the actual distributions used are in the legacy directory. Fit breakpoints accordingly in numpy/numpy/lib/random/src/legacy/legacy-distributions.c.