Debugging Techniques

pdb Fundamentals

Drop into an interactive session with the builtin (Python 3.7+):

def total(rows):
    s = 0
    for r in rows:
        breakpoint()          # pauses here, no import needed
        s += r["amount"]
    return s

Commands at the (Pdb) prompt:

Post-mortem on an exception: python -m pdb script.py, then c — pdb stops at the raise point with full locals.

Breakpoints, ipdb, and Tests

Environment variables redirect or silence the builtin:

PYTHONBREAKPOINT=0 python app.py            # disable all breakpoint() calls
PYTHONBREAKPOINT=ipdb.set_trace python app.py  # richer REPL with completion

Set conditional breakpoints in code (if attempt == 3: breakpoint()) or from the pdb prompt with b file.py:42, condition / b func, expr.

For tests, pytest --pdb opens the debugger at the first failing assert; pair it with --tb=short for readable output. Watch one habit: inspect with p, then continue — mutating objects while exploring silently changes program state and can hide the very bug you are chasing.

Pitfalls and Interview Angles

Print-debugging fails for concurrency. Interleaved async tasks or threads scramble print order; logging with task IDs or a debugger with per-frame inspection stays coherent.

Read tracebacks bottom-up. The last line is the actual error; frames above show the call chain. Chained exceptions (raise X from Y) link causes — inspect __cause__ instead of guessing.

Debugger vs production. Breakpoints freeze one process; in production use structured logging, faulthandler, or remote profilers.

Interview angle: describe a debugging loop — reproduce deterministically, form a hypothesis, place one breakpoint or log line to test it, confirm, remove the probe — and name the pdb command you would use to inspect a value in a parent frame (w, then u, then p).