beginner25 minby DevFox

When Things Go Wrong: Exceptions

Reading a traceback, catching only what you can handle, and why except Exception hides the bug you most needed to see. try, except, else and finally, and raising your own.

Python

Your photo script has worked perfectly for a week. This morning someone put a file called notes.txt in the photos folder, and the whole thing stopped on photo eleven of four hundred — with a wall of red text and no indication of which photos had been processed and which had not.

Every program meets things it did not expect. The question is not how to prevent that, because you cannot; it is what your program does when it happens. By the end of this lesson you will be able to read the red text, decide what to catch and what to let through, and — the part people get wrong — know why catching everything is worse than catching nothing.

What an exception is

When Python cannot do what you asked, it raises an exception: it stops what it was doing and looks for someone who has said they will handle this situation. If nobody has, the program stops and prints a traceback.

python

int("four")
# ValueError: invalid literal for int() with base 10: 'four'

That is not a crash in the sense of something going wrong with Python. It is Python working correctly — refusing to guess, and telling you exactly what it could not do.

The ones you will meet first:

text

ValueError        right type, impossible value      int("four")
TypeError         wrong type entirely               "a" + 5
KeyError          no such key in a dictionary       d["missing"]
IndexError        position beyond the end           photos[999]
FileNotFoundError no file at that path              open("nope.txt")
ZeroDivisionError dividing by zero                  10 / 0
AttributeError    no such method or attribute       "text".appendx()
NameError         no such name                      print(phtoo)

Reading them is a skill worth ten minutes. AttributeError: 'NoneType' object has no attribute 'name' looks cryptic and says something very specific: something you expected to be an object is None, and the real bug is wherever that None came from — usually a function that returned nothing, several lines earlier.

Handling one

try and except say "attempt this; if that particular thing goes wrong, do this instead":

python

try:
    count = int(input("How many photos? "))
except ValueError:
    print("That was not a number. Using 100.")
    count = 100

Everything in the try block runs normally until something raises. If what raises matches the except, the handler runs and the program continues. If it does not match, the exception carries on upward as though there were no try at all.

You can handle several kinds:

python

try:
    photo = load(path)
except FileNotFoundError:
    print(f"No file at {path}")
except PermissionError:
    print(f"Cannot read {path} - check permissions")

Or treat several the same way:

python

except (FileNotFoundError, PermissionError) as error:
    print(f"Could not read {path}: {error}")

as error gives you the exception object, which prints as its message and carries details. Including it costs nothing and is the difference between a log line that helps and one that says "something went wrong".

Catch what you can handle

Now the rule the whole lesson turns on.

Bad — catching everything and carrying on.

python

for path in paths:
    try:
        process(load(path))
    except Exception:
        continue

Good — catching the expected failure, letting the rest through.

The first version does what it was written to do — one bad file does not stop the run. It also swallows every other possible problem: the typo in process that raises AttributeError, the disk filling up, the bug that raises KeyError on every single photo. The script reports success, the output folder is empty, and there is no record anywhere of why. except Exception: continue is how a program lies to you.

The second catches the two failures it genuinely knows how to respond to and says so out loud. Anything else stops the program and prints a traceback — which is the correct outcome for a problem nobody anticipated, because a loud failure gets fixed and a silent one does not.

The principle: catch the exception you have a plan for. If your handler is pass, or continue, or a comment saying "this shouldn't happen", you did not have a plan.

The other two blocks

try takes two more clauses, and both earn their place. All four together read in order.

try — the one line that might fail
Keep it small. A fat try block catches failures from code you were not worried about and blames them on the code you were.
except — only if it raised
And only if the kind matches. Anything else carries on upward as though the try were not there.
else — only if it did not raise
The follow-on work. Putting it here rather than inside the try is how you keep the try block down to one line.
finally — either way, always
Success, exception, even a return from inside the try. This is the cleanup guarantee.

Keeping each block to its own job is what stops a handler catching something it was not meant to

else runs only if nothing was raised:

Putting process(photo) here rather than inside the try keeps the try block down to the one line that might fail. That matters: if process happened to raise FileNotFoundError for its own reasons, a fat try block would catch it and blame the load.

finally runs either way, raised or not:

That is the cleanup guarantee. Whatever happens — success, an exception, even a return from inside the try — the finally block runs. For files and connections there is a tidier tool (with, in the next lesson), but finally is what it is built on.

Raising your own

Your code can raise exceptions too, and should. When a function is given something it cannot work with, saying so immediately is kinder than continuing and producing nonsense:

Include the offending value in the message. "width must be positive" sends the reader looking; "got -50" often tells them what is wrong on its own.

For situations specific to your program, define your own type:

That is a complete definition — you will meet class properly in a later lesson, and this is one of the few places you need it before then. The benefit is that callers can catch this specifically:

With a plain ValueError they would have to catch every ValueError in the call, including ones from unrelated code.

Asking permission or asking forgiveness

Two styles solve the same problem, and Python has a preference.

Python leans toward the second, and there is a concrete reason beyond taste.

Check first

Fine when the test is cheap and local.

if key in dictionary is clearer than catching a KeyError, and nothing else can change that dictionary between the two lines.

Against a file or a network, the same shape is a trap: the world can change in the gap, and a check that something else can invalidate is not a check.

Try, and handle failure

No gap. The attempt is the check.

The file cannot be deleted between the test and the use, because there is no separate test.

This is why Python code reads the way it does around files, networks and anything else outside the program.

Check for things you control. Catch for things you do not.

A cheat sheet to keep

Where to go next

You can now read a traceback, handle the failures you expect, raise informative errors of your own, and guarantee cleanup runs. The judgement to keep is about scope: catch narrowly, say something when you do, and let the unexpected reach you loudly. A program that fails clearly is worth far more than one that appears to succeed.

Next is Reading and Writing Files, where most of these exceptions actually come from. It introduces with, which is finally in a form you will use every day, and the encoding questions that turn text on disk into text in your program.

Before you move on, take a program you have written and feed it something wrong on purpose — a missing file, letters where it wanted digits. Read each traceback, then handle exactly that one exception and leave the others alone. Then try except Exception: pass and watch a real bug disappear without trace. The second experiment is the one that makes the rule stick.