We Keep Coding

Python: Best practice to report an error during closing a resource

What is the best practice in Python for reporting an error which occurs when closing a resource?
In particular, if I implement __enter__ and __exit__, I can use
with my_library.connect(endpoint) as connection:
connection.store_data("...")
But what should I do if closing my connection and persisting the changes fails, e.g. due to a network outage? I know that I could technically raise an error from within __exit__ but is this best practice/idiomatic in Python? Or should I, e.g., provide a separate persistChanges method, let the __exit__ swallow all errors, and then write in the documentation "if you don't call persistChanges you might lose your changes in error cases"?
My specific use case is: I am providing a Python API to other developers, and I wonder how to handle this case of "error on closing the resource" such that my API follows Python best practices/meets the expectations of Python devs using my library.

I would recommend making a custom error/warning class for your library. this can be very, very simple. There already exists a set of built-in exceptions that you can extend from. Based on your description above, I would recommend extending the RuntimeError like this:
class MyLibraryConnectionError(RuntimeError):
pass
or, if you want to only throw a warning, using the ResourceWarning like this:
class MyLibraryConnectionWarning(ResourceWarning):
pass
There is also the RuntimeWarning that could be extended for similar effect.
If you feel that ResourceWarning, RuntimeWarning, and RuntimeError don't accurately describe the exception, you can also just have them inherit directly from Exception or Warning, depending on whether you want them to only be flagged in Developer mode(Warning), or if you want the full exception functionality.
You can throw these like any other exception:
throw MyLibraryConnectionError("The underlying resource failed to close")
throw MyLibraryConnectionWarning("The underlying resource failed to close")
or even catch your dependency's thrown exception:
def __exit__(...):
try:
# potentially dangerous code with connections
underlyingConnection.close()
except TheErrorYouSeeThrown as e: # you can probably make this Exception instead of TheErrorYouSeeThrown. The more specific the better, so you don't accidentally catch random errors you didn't mean to.
throw MyLibraryConnectionError(e.message) # or whatever this needs to be, depending on the thrown exception type
Then your users could implement like so:
try:
with my_library.connect(endpoint) as connection:
connection.store_data("...")
except my_library.MyLibraryConnectionError as e:
# handle appropriately, or ignore
except Exception as e:
# Handle other errors that happen, that your library doesn't cause.

Is there a use case for an alternative "exception cause" in Python?

Background
In Python it is possible to suppress the context of an Exception if you raise your Exception from None. PEP 409 describes the rationale behind it. Sometimes you want to show only one meaningful (custom) Exception. With PEP 415 the implementation changes with the following argument:
The main problem with this scheme [from PEP 409] is it complicates the role of __cause__. __cause__ should indicate the cause of the exception not whether __context__ should be printed or not. This use of __cause__ is also not easily extended in the future. For example, we may someday want to allow the programmer to select which of __context__ and __cause__ will be printed.
Question
PEP 419 talks about future use cases. Are there any valid use cases right now (Python 3.3+) for using an alternative exception cause? For example, consider the following code:
class CustomError(BaseException):
pass
class StupidError(BaseException):
def __init__(self, message='This is just a stupid error.'):
super(StupidError, self).__init__(message)
self.message = message
try:
value = int('a')
except Exception:
raise CustomError('Custom error message') from StupidError
Output:
StupidError: This is just a stupid error.
The above exception was the direct cause of the following exception:
Traceback (most recent call last): (...)
CustomError: Custom error message
Are there any real use cases in which you want to hide the ValueError but show the StupidError? I mean you could give some relevant information in the StupidError which are not present in a mere ValueError? Maybe I am just overthinking this whole thing.

Sure. Say you're writing a web app and need to hide sensitive/complicated data from the user.
Say a user tries a request, but the web app backend has trouble reading the necessary data from a MySQL database (for whatever reason). Instead of letting the MySQL module raise the error, which could either expose sensitive information about how the app works internally or just simply confuse the user, I would want to catch it and then throw my custom exception (let's call it serverError). That custom exception would show the user a HTTP 500 page as well as report the error internally so it can be analyzed by a developer to figure out what went wrong and how to prevent it.
This means I only have to write my general error handling code once, and then when I catch an error, I can raise serverError, which takes care of the error reporting for me.

Catching Python Exceptions and printing out a separate message

I'm currently trying to write code to catch exceptions, and depending upon which exception is thrown, a different module will be imported than when no exception was thrown.
try:
import sge_execution_engine as execution_engine
except ImportError:
print "Use local execution engine because SGE support is missing!"
print sys.exc_info() # Print out the exception message
import local_execution_engine as execution_engine
except RuntimeError:
print "Using local execution engine because SGE support is missing!"
print sys.exc_info()
import local_execution_engine as execution_engine
The first exception, ImportError that is caught, catches the exception thrown when python drmaa module cannot be found during the execution of import sge_execution_engine (inside sge_execution_engine, there is an import drmaa statement). The second exception, RuntimeError, is caught when the drmaa python library is found (likewsie during the execution of the import drmaa statement inside the sge_execution_engine), but the drmaa C library is not installed into the OS. We hope that these two except statements are sufficient to catch all possible exceptions that can be thrown when a user attempts to run this module on a machine that just does not have the python drmaa library, the drmaa C library, or does not have Sun Grid Engine installed. without any of these proceeds, the module proceeds to then import local_execution_engine and so the code can then execute on the user's machine locally. Right now the code works as expected in the sense that it goes to import local when it finds exceptions with sge, but we are still looking to improve the exception handling here to make it more robust.
In my opinion I think having the actual Exception message that was thrown be printed to stdout is good as it will allow the user to know why he was unable to import sge_execution_engine especially if he was not expecting it to fail being imported.
However, instead of using print sys.exc_info() to actually have the actual exception message be printed on screen, I realized that perhaps a better way would be to use the except EXCEPTION as some_variable_name format and then print out print some_variable_name and also call some of the attributes associated with the Exception that is thrown and assigned to some_variable_name.
I saw this being done in the Python tutorial on exceptions where there was this chunk of code:
import sys
try:
f = open('myfile.txt')
s = f.readline()
i = int(s.strip())
except IOError as e:
print "I/O error({0}): {1}".format(e.errno, e.strerror)
except ValueError:
print "Could not convert data to an integer."
except:
print "Unexpected error:", sys.exc_info()[0]
raise
It seems like the except IOError as e chunk is handling the exception message in a fine-grained way by specifically calling on the errno and strerror attributes of the IOError object. However, when I look at the IOError documentation , I do not see these specific attributes being listed out as part of the documentation for the exception. In fact, this is also the case for all the other exceptions under the Python documentation, so it seems there is no way we can figure out what attributes will be associated with a particular exception. If we don't know anything about this, then how will we be able to figure out what attributes to call on the some_variable_name object when we use the import EXCEPTION as some_variable_name syntax to handle our exceptions?
I would appreciate anyone's suggestion on this, and even if your answer is not directly answering my question, but if you have another entirely different suggestion on how I could better handle my exception here, please don't hesitate to make a post!
Thank you very much!

Using a bare except is rarely a good idea, but this is one of those rare times -- primarily because you have a backup that you want to use if, for any reason, you cannot import the system sge:
try:
import sge_execution_engine as execution_engine
except:
print "Use local execution engine because SGE support is missing!"
print sys.exc_info() # Print out the exception message
import local_execution_engine as execution_engine
Notice you now only need one except clause.
The better way to handle getting that information to the user is probably to go ahead and print it out, and then also use the logging module to make a permanent record:
import logging
logger = logging.getLogger()
logger.setLevel(logging.WARNING)
and then in your except clause add:
logger.exception('unable to import sge')
and your message, along with the actual exception, will be saved in a log file.

First you're right that it's better to catch the exception into a variable than to ignore it and then pull it back out with sys.exc_info(). There are a few good reasons to use exc_info (low-level code, code that has to work with both pre-2.6 and 3.x, etc.), but in general, when you get can do it your way, you should.
And this works even for your last bare except. In 2.7, a plain except: means the same thing as except Exception:, so you can write except Exception as e:, and then use the e value.
Also note that if you want to do exactly the same thing for multiple exception types, you can write except (RuntimeError, ImportError) as e:.
As for "making it more robust", that's not an absolute thing. For example, if there's some unexpected exception that's neither a RuntimeError nor an ImportError, do you want to log it and try the fallback code, or quite and dump the traceback? If it's, e.g., a SyntaxError caused by someone checking in a bad edit to your program, you may not want to treat that like a runtime error… or maybe you do; that really depends on your development practices and target userbase.
Meanwhile:
It seems like the except IOError as e chunk is handling the exception message in a fine-grained way by specifically calling on the errno and strerror attributes of the IOError object. However, when I look at the IOError documentation, I do not see these specific attributes being listed out as part of the documentation for the exception.
You need to look up the hierarchy. Notice that IOError is a subclass of EnvironmentError, which does document the errno and strerror attributes. (What these attributes actually mean is only documneted for OSError and its subclasses, but the fact that they exist is documented.)
If you think this is all a bit of a mess… well, it is. It's all cleaned up in Python 3.x, where IOError and EnvironmentError are merged into OSError, which clearly documents its attributes, and where you usually don't have to switch on errno in the first place because common errno values generate a specific subclass like FileNotFoundError, and so on. But as long as you're using 2.7, you don't get the benefits of the last 6 years of improvements to the language.
For example, looking at the hierarchy or ValueError=>StandardError=>Exception (from lowest to highest in the hierarchy), I can't find any attributes about it.
If you dir a ValueError, you'll see that it only has two attributes (besides the usual special stuff like __repr__ and __class_): args and message.
message isn't documented because it was deprecated in 2.5, and only exists in 2.7 to allow some pre-2.5 code to keep running.* But args is documented; you just need to go up one level further, to BaseException:
args
The tuple of arguments given to the exception constructor. Some built-in exceptions (like IOError) expect a certain number of arguments and assign a special meaning to the elements of this tuple, while others are usually called only with a single string giving an error message.
So, the reason you can't find the other attributes in ValueError is that there are no other attributes to find. And the same goes for those other classes. The handful of types that have special attributes (OSError, SyntaxError, maybe a few module-specific types elsewhere in the stdlib) document them explicitly.**
If we are using the except some_exception as e syntax, is doing a print e sufficient to get the exception printed out without calling its attributes
It's sufficient to get some useful form of the exception printed. Again, from the BaseException docs:
If str() or unicode() is called on an instance of this class, the representation of the argument(s) to the instance are returned or the empty string when there were no arguments.
In some cases, that's not what you want. In particular, notice that it doesn't include the type of the exception. You can get that with repr, which gives you something that looks like a constructor call (e.g., ValueError("invalid literal for int() with base 10: 'f'")).
If you want the same output you get in a traceback, you have to put the type and the str or unicode together yourself—e.g., '{}: {}'.format(type(e), e).
If you want to get the actual information out of the exception, like that base 10 or that string 'f'—well, you can't,*** because that information has already been thrown away. You'll have to write your own code to keep track of it, like:
try:
i = int(s, 16)
except ValueError as e:
print '{} is not a base-16 number'.format(s)
* It's as if BaseException.__init__(self, *args) were defined to do self.args = tuple(args); self.message = str(args[0]) if args else ''.
** I believe in 2.5 there were a few exception types that had undocumented attributes as an implementation detail of CPython, but by 2.7 they're all either gone or documented.
*** Well, you could parse the exception string, but needless to say, that's an implementation detail, not something guaranteed to be stable and portable. It could be different in Jython, or on a Spanish-language system, or it may not quote strings the way you expect, etc.

Catch response code with Python 2.6 unittest

Using unittest to test some edge cases with an API. All of these cases return 400-series response codes using custom exception classes. Unfortunately, I can't seem to find a way to catch these custom exceptions or to read the response codes in the unittest check itself.
What I have been receiving is an 'AppError' exception with a message mentioning it is a 400 response rather than a 200 or 300. I want to avoid parsing the exception message if I can. This also needs to work on Python 2.6+. How can I either catch the custom exceptions in my unittest check, or determine the response code causing the error without parsing the exception message?

You'll have to catch the exception and assert that it is not a 400 response:
try:
call_api_method()
except AppError as ae:
self.assert(ae.errorcode < 400)
This requires the exception to carry the error code as an attribute; you perhaps need to inspect what attributes are available on the exception. By default .args will be there (it is a tuple), but it is good practice for such exceptions to have an error code attribute too.

I figured out how to do this. Turns out that with webtest (which I forgot to mention as being used), you can pass in a parameter expect_errors=True and save the return value of the request. You can then check against the response status with
self.assertEqual('400 Bad Request', response.status)

How Can I Find a List of All Exceptions That a Given Library Function Throws in Python?

Sorry for the long title, but it seems most descriptive for my question.
Basically, I'm having a difficult time finding exception information in the official python documentation. For example, in one program I'm currently writing, I'm using the shutil libary's move function:
from shutil import move
move('somefile.txt', '/tmp/somefile.txt')
That works fine, as long as I have write access to /tmp/, there is enough diskspace, and if all other requirements are satisfied.
However, when writing generic code, it is often difficult to guarantee those factors, so one usually uses exceptions:
from shutil import move
try:
move('somefile.txt', '/tmp/somefile.txt')
except:
print 'Move failed for some reason.'
I'd like to actually catch the appropriate exceptions thrown instead of just catching everything, but I simply can't find a list of exceptions thrown for most python modules. Is there a way for me to see which exceptions a given function can throw, and why? This way I can make appropriate cases for each exception, eg:
from shutil import move
try:
move('somefile.txt', '/tmp/somefile.txt')
except PermissionDenied:
print 'No permission.'
except DestinationDoesNotExist:
print "/tmp/ doesn't exist"
except NoDiskSpace:
print 'No diskspace available.'
Answer points go to whoever can either link me to some relevant documentation that I've somehow overlooked in the official docs, or provide a sure-fire way to figure out exactly which exceptions are thrown by which functions, and why.
Thanks!
UPDATE: It seems from the answers given that there really isn't a 100% straight-forward way to figure out which errors are thrown by specific functions. With meta programming, it seems that I can figure out simple cases and list some exceptions, but this is not a particularly useful or convenient method.
I'd like to think that eventually there will be some standard for defining which exceptions are raised by each python function, and that this information will be included in the official documentation. Until then I think I will just allow those exceptions to pass through and error out for my users as it seems like the most sane thing to do.

To amplify Messa, catch what you expect are failure modes that you know how to recover from. Ian Bicking wrote an article that addresses some of the overarching principles as does Eli Bendersky's note.
The problem with the sample code is that it is not handling errors, just prettifying them and discarding them. Your code does not "know" what to do with a NameError and there isn't much it should do other than pass it up, look at Bicking's re-raise if you feel you must add detail.
IOError and OSError are reasonably "expectable" for a shutil.move but not necessarily handleable. And the caller of your function wanted it to move a file and may itself break if that "contract" that Eli writes of is broken.
Catch what you can fix, adorn and re-raise what you expect but can't fix, and let the caller deal with what you didn't expect, even if the code that "deals" is seven levels up the stack in main.

Python doesn't have a mechanism right now for declaring which exceptions are thrown, unlike (for example) Java. (In Java you have to define exactly which exceptions are thrown by what, and if one of your utility methods needs to throw another exception then you need to add it to all of the methods which call it which gets boring quickly!)
So if you want to discover exactly which exceptions are thrown by any given bit of python then you need to examine the documentation and the source.
However python has a really good exception hierarchy.
If you study the exception hierarchy below you'll see that the error superclass you want to catch is called StandardError - this should catch all the errors that might be generated in normal operations. Turning the error into into a string will give a reasonable idea to the user as to what went wrong, so I'd suggest your code above should look like
from shutil import move
try:
move('somefile.txt', '/tmp/somefile.txt')
except StandardError, e:
print 'Move failed: %s' % e
Exception hierarchy
BaseException
|---Exception
|---|---StandardError
|---|---|---ArithmeticError
|---|---|---|---FloatingPointError
|---|---|---|---OverflowError
|---|---|---|---ZeroDivisionError
|---|---|---AssertionError
|---|---|---AttributeError
|---|---|---BufferError
|---|---|---EOFError
|---|---|---EnvironmentError
|---|---|---|---IOError
|---|---|---|---OSError
|---|---|---ImportError
|---|---|---LookupError
|---|---|---|---IndexError
|---|---|---|---KeyError
|---|---|---MemoryError
|---|---|---NameError
|---|---|---|---UnboundLocalError
|---|---|---ReferenceError
|---|---|---RuntimeError
|---|---|---|---NotImplementedError
|---|---|---SyntaxError
|---|---|---|---IndentationError
|---|---|---|---|---TabError
|---|---|---SystemError
|---|---|---TypeError
|---|---|---ValueError
|---|---|---|---UnicodeError
|---|---|---|---|---UnicodeDecodeError
|---|---|---|---|---UnicodeEncodeError
|---|---|---|---|---UnicodeTranslateError
|---|---StopIteration
|---|---Warning
|---|---|---BytesWarning
|---|---|---DeprecationWarning
|---|---|---FutureWarning
|---|---|---ImportWarning
|---|---|---PendingDeprecationWarning
|---|---|---RuntimeWarning
|---|---|---SyntaxWarning
|---|---|---UnicodeWarning
|---|---|---UserWarning
|---GeneratorExit
|---KeyboardInterrupt
|---SystemExit
This also means that when defining your own exceptions you should base them off StandardError not Exception.
Base class for all standard Python exceptions that do not represent
interpreter exiting.

Yes, you can (for simple cases), but you need a bit of meta-programming. Like the other answers have said, a function does not declare that it throws a particular error type, so you need to look at the module and see what exception types it defines, or what exception types it raises. You can either try to grok the documentation or leverage the Python API to do this.
To first find which exception types a module defines, just write a simple script to go through each object in the module dictionary module.__dict__ and see if it ends in the word "Error" or if it is a subclass of Exception:
def listexns(mod):
"""Saved as: http://gist.github.com/402861
"""
module = __import__(mod)
exns = []
for name in module.__dict__:
if (issubclass(module.__dict__[name], Exception) or
name.endswith('Error')):
exns.append(name)
for name in exns:
print '%s.%s is an exception type' % (str(mod), name)
return
If I run this on your example of shutils I get this:
$ python listexn.py shutil
Looking for exception types in module: shutil
shutil.Error is an exception type
shutil.WindowsError is an exception type
$
That tells you which error types are defined, but not which ones are thrown. To achieve the latter, we need to walk over the abstract syntax tree generated when the Python interpreter parses the module, and look for every raise statement, then save a list of names which are raised. The code for this is a little long, so first I'll state the output:
$ python listexn-raised.py /usr/lib/python2.6/shutil.py
Looking for exception types in: /usr/lib/python2.6/shutil.py
/usr/lib/python2.6/shutil.py:OSError is an exception type
/usr/lib/python2.6/shutil.py:Error is an exception type
$
So, now we know that shutil.py defines the error types Error and WindowsError and raises the exception types OSError and Error. If we want to be a bit more complete, we could write another method to check every except clause to also see which exceptions shutil handles.
Here's the code to walk over the AST, it just uses the compiler.visitor interface to create a walker which implements the "visitor pattern" from the Gang of Four book:
class ExceptionFinder(visitor.ASTVisitor):
"""List all exceptions raised by a module.
Saved as: http://gist.github.com/402869
"""
def __init__(self, filename):
visitor.ASTVisitor.__init__(self)
self.filename = filename
self.exns = set()
return
def __visitName(self, node):
"""Should not be called by generic visit, otherwise every name
will be reported as an exception type.
"""
self.exns.add(node.name)
return
def __visitCallFunc(self, node):
"""Should not be called by generic visit, otherwise every name
will be reported as an exception type.
"""
self.__visitName(node.node)
return
def visitRaise(self, node):
"""Visit a raise statement.
Cheat the default dispatcher.
"""
if issubclass(node.expr1, compiler.ast.Name):
self.__visitName(node.expr1)
elif isinstance(node.expr1, compiler.ast.CallFunc):
self.__visitCallFunc(node.expr1)
return

As these operations usually use libc functions and operating system calls, mostly you get IOError or OSError with an errno number; these errors are listed in man pages of that libc/OS calls.
I know this is possibly not a complete answer, it would be good to have all exceptions listed in documentation...