Python exception chaining

Question

Is there a standard way of using exception chains in Python? Like the Java exception 'caused by'?

Here is some background.

I have a module with one main exception class DSError:

 class DSError(Exception):
     pass

Somewhere within this module there will be:

try:
    v = my_dict[k]
    something(v)
except KeyError as e:
    raise DSError("no key %s found for %s" % (k, self))
except ValueError as e:
    raise DSError("Bad Value %s found for %s" % (v, self))
except DSError as e:
    raise DSError("%s raised in %s" % (e, self))

Basically this snippet should throw only DSError and tell me what happened and why. The thing is that the try block might throw lots of other exceptions, so I'd prefer if I can do something like:

try:
    v = my_dict[k]
    something(v)
except Exception as e:
    raise DSError(self, v, e)  # Exception chained...

Is this standard pythonic way? I did not see exception chains in other modules so how is that done in Python?

Solution 1

Exception chaining is only available in Python 3, where you can write:

try:
    v = {}['a']
except KeyError as e:
    raise ValueError('failed') from e

which yields an output like

Traceback (most recent call last):
  File "t.py", line 2, in <module>
    v = {}['a']
KeyError: 'a'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "t.py", line 4, in <module>
    raise ValueError('failed') from e
ValueError: failed

In most cases, you don't even need the from; Python 3 will by default show all exceptions that occured during exception handling, like this:

Traceback (most recent call last):
  File "t.py", line 2, in <module>
    v = {}['a']
KeyError: 'a'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "t.py", line 4, in <module>
    raise ValueError('failed')
ValueError: failed

What you can do in Python 2 is adding custom attributes to your exception class, like:

class MyError(Exception):
    def __init__(self, message, cause):
        super(MyError, self).__init__(message + u', caused by ' + repr(cause))
        self.cause = cause

try:
    v = {}['a']
except KeyError as e:
    raise MyError('failed', e)

Solution 2

Is this what you're asking for?

class MyError(Exception):
    def __init__(self, other):
        super(MyError, self).__init__(other.message)

>>> try:
...     1/0
... except Exception, e:
...     raise MyError(e)
Traceback (most recent call last):
  File "<pyshell#27>", line 4, in <module>
    raise MyError(e)
MyError: division by zero

If you want to store the original exception object, you can certainly do so in your own exception class's __init__. You might actually want to store the traceback as the exception object itself doesn't provide much useful information about where the exception occurred:

class MyError(Exception):
    def __init__(self, other):
        self.traceback = sys.exc_info()
        super(MyError, self).__init__(other.message)

After this you can access the traceback attribute of your exception to get info about the original exception. (Python 3 already provides this as the __traceback__ attribute of an exception object.)

Comments

[Ayman]

Almost right, but I considered this would be 'cheating', since it only takes the message of the chained exception, not the actual exception object. I.e. I would not know where the actual division by zero occurred, just that it was caught somewhere.

[BrenBarn]

@Ayman: See my edited answer. All you have to do is grab the traceback and store it. However, if you really want all the information from the original exception to show up in the traceback like a real exception, then phihag is right that this can't be accomplished in Python 2. You'd have to just manually print the old traceback as part of your exception's message.

[Ayman]

Thanks. I didn't know about the sys.exc_info(). I would accept this as the answer too :-)

[Mr_and_Mrs_D]

is other.message always present ?

[Mr_and_Mrs_D]

For python 2 if one wants to save the traceback - which one must want - raise MyError(message + u', caused by ' + repr(cause)), None, sys.exc_info()[2]

[timgeb]

"In most cases, you don't even need the from" Do you have an example where it is needed or useful?

[Eric Smith]

@timgeb PEP 3134 has two situations for chaining: one where error handling code results in another exception being raised, and the other where an exception was deliberately translated to a different exception. The from e is for the deliberate case, and changes the message in the output as shown in the answer above.

[Alvaro Gutierrez Perez]

This response is, in my opinion, better than the duplicated question's selected answer, as this covers Python 3. Also, kudos for noting that the from isn't even necessary, as Python 3 tracebacks already print all current exceptions in the stack.

[PirateApp]

upvoted! what would you do if you also want to return a value from this

[phihag]

@PirateApp You can either raise or return, can't do both. But if you need a value, you can simply attach it to the exception object, for instance by making the value a property of your custom error class (MyError here).

[variable]

So in python 3 chaining is default behaviour? And only when deliberately raising another exception of we want to get full trace then we use the raise from keyword? So are raise ex, raise, raise from all the same?