Debugging Strategies

Why this matters

Programmers spend more time fixing code than writing it. The skill is not being lucky; it is a method. A good habit turns a mysterious failure into a short, boring search.

The idea

1. Reproduce it. Find the smallest input that fails: last_n([1, 2, 3], 0).
2. Read the message. The last line of a traceback names the error; the line above points at your code.
3. Look at the values. Print what is really there, not what you think is there. f"{total=}" prints both the name and the value:

   total = 7
   print(f"{total=}")      # total=7

4. Shrink the problem. Cut code away (or add print calls halfway) until you know which half is wrong.
5. Check your assumptions. assert len(items) > 0, "items is empty" stops the program right where an assumption breaks.
6. Explain it aloud (the “rubber duck”). Saying what each line does often shows where it does not.

Most bugs are one of a handful of kinds: off-by-one, an edge case (empty, zero, one item, capitals), the wrong variable, or changing a list while looping over it.

Try it

Predict what last_n([1, 2, 3, 4], 2) returns, then run it with a print inside. Then try n = 0.

Your turn

The starter contains four functions. Each one has a bug. Fix all four without changing what they are supposed to do:

• average(numbers): the mean of a list of numbers.
• count_vowels(text): how many vowels (a e i o u, either case) are in the text.
• last_n(items, n): the last n items as a new list ([] when n is 0 or less; the whole list if n is bigger than its length).
• remove_negatives(numbers): a list without the negative numbers.

The test names and messages tell you what input failed. Reproduce, look, fix.

Recap

• Reproduce with the smallest failing input, then look at real values.
• Use print/f"{x=}" and assert to check your assumptions.
• Suspect edge cases and lists being changed while you loop over them.

Common mistakes

• Changing several things at once. Change one thing, run again.
• Trusting the code instead of the output. Print it.
• Fixing the symptom (special-casing one input) instead of the cause.