|Title:||Exception Reorganization for Python 3.0|
|Author:||Brett Cannon <brett at python.org>|
- Rationale For Wanting Change
- Philosophy of Reorganization
- New Hierarchy
- Differences Compared to Python 2.4
- Required Superclass for raise
- Bare except Clauses Catch Exception
- Transition Plan
- Rejected Ideas
- DeprecationWarning Inheriting From PendingDeprecationWarning
- AttributeError Inheriting From TypeError or NameError
- Removal of EnvironmentError
- Introduction of MacError and UnixError
- SystemError Subclassing SystemExit
- ControlFlowException Under Exception
- Rename NameError to NamespaceError
- Renaming Existing Exceptions
- Have EOFError Subclass IOError
- Have MemoryError and SystemError Have a Common Superclass
- Common Superclass for PendingDeprecationWarning and DeprecationWarning
- Removing WindowsError
- Superclass for KeyboardInterrupt and SystemExit
This PEP has been rejected .
Python, as of version 2.4, has 38 exceptions (including warnings) in the built-in namespace in a rather shallow hierarchy. These classes have come about over the years without a chance to learn from experience. This PEP proposes doing a reorganization of the hierarchy for Python 3.0 when backwards-compatibility is not as much of an issue.
Along with this reorganization, adding a requirement that all objects passed to a raise statement must inherit from a specific superclass is proposed. This is to have guarantees about the basic interface of exceptions and to further enhance the natural hierarchy of exceptions.
Lastly, bare except clauses will be changed to be semantically equivalent to except Exception. Most people currently use bare except clause for this purpose and with the exception hierarchy reorganization becomes a viable default.
Exceptions are a critical part of Python. While exceptions are traditionally used to signal errors in a program, they have also grown to be used for flow control for things such as iterators.
While their importance is great, there is a lack of structure to them. This stems from the fact that any object can be raised as an exception. Because of this you have no guarantee in terms of what kind of object will be raised, destroying any possible hierarchy raised objects might adhere to.
But exceptions do have a hierarchy, showing the severity of the exception. The hierarchy also groups related exceptions together to simplify catching them in except clauses. To allow people to be able to rely on this hierarchy, a common superclass that all raise objects must inherit from is being proposed. It also allows guarantees about the interface to raised objects to be made (see PEP 344). A discussion about all of this has occurred before on python-dev .
As bare except clauses stand now, they catch all exceptions. While this can be handy, it is rather overreaching for the common case. Thanks to having a required superclass, catching all exceptions is as easy as catching just one specific exception. This allows bare except clauses to be used for a more useful purpose. Once again, this has been discussed on python-dev .
Finally, slight changes to the exception hierarchy will make it much more reasonable in terms of structure. By minor rearranging exceptions that should not typically be caught can be allowed to propagate to the top of the execution stack, terminating the interpreter as intended.
For the reorganization of the hierarchy, there was a general philosophy followed that developed from discussion of earlier drafts of this PEP , , , , , . First and foremost was to not break anything that works. This meant that renaming exceptions was out of the question unless the name was deemed severely bad. This also meant no removal of exceptions unless they were viewed as truly misplaced. The introduction of new exceptions were only done in situations where there might be a use for catching a superclass of a category of exceptions. Lastly, existing exceptions would have their inheritance tree changed only if it was felt they were truly misplaced to begin with.
For all new exceptions, the proper suffix had to be chosen. For those that signal an error, "Error" is to be used. If the exception is a warning, then "Warning". "Exception" is to be used when none of the other suffixes are proper to use and no specific suffix is a better fit.
After that it came down to choosing which exceptions should and should not inherit from Exception. This was for the purpose of making bare except clauses more useful.
Lastly, the entire existing hierarchy had to inherit from the new exception meant to act as the required superclass for all exceptions to inherit from.
Exceptions flagged with "stricter inheritance" will no longer inherit from a certain class. A "broader inheritance" flag means a class has been added to the exception's inheritance tree. All comparisons are against the Python 2.4 exception hierarchy.
+-- BaseException (new; broader inheritance for subclasses) +-- Exception +-- GeneratorExit (defined in PEP 342) +-- StandardError +-- ArithmeticError +-- DivideByZeroError +-- FloatingPointError +-- OverflowError +-- AssertionError +-- AttributeError +-- EnvironmentError +-- IOError +-- EOFError +-- OSError +-- ImportError +-- LookupError +-- IndexError +-- KeyError +-- MemoryError +-- NameError +-- UnboundLocalError +-- NotImplementedError (stricter inheritance) +-- SyntaxError +-- IndentationError +-- TabError +-- TypeError +-- RuntimeError +-- UnicodeError +-- UnicodeDecodeError +-- UnicodeEncodeError +-- UnicodeTranslateError +-- ValueError +-- ReferenceError +-- StopIteration +-- SystemError +-- Warning +-- DeprecationWarning +-- FutureWarning +-- PendingDeprecationWarning +-- RuntimeWarning +-- SyntaxWarning +-- UserWarning + -- WindowsError +-- KeyboardInterrupt (stricter inheritance) +-- SystemExit (stricter inheritance)
A more thorough explanation of terms is needed when discussing inheritance changes. Inheritance changes result in either broader or more restrictive inheritance. "Broader" is when a class has an inheritance tree like cls, A and then becomes cls, B, A. "Stricter" is the reverse.
The superclass that all exceptions must inherit from. It's name was chosen to reflect that it is at the base of the exception hierarchy while being an exception itself. "Raisable" was considered as a name, it was passed on because its name did not properly reflect the fact that it is an exception itself.
Direct inheritance of BaseException is not expected, and will be discouraged for the general case. Most user-defined exceptions should inherit from Exception instead. This allows catching Exception to continue to work in the common case of catching all exceptions that should be caught. Direct inheritance of BaseException should only be done in cases where an entirely new category of exception is desired.
But, for cases where all exceptions should be caught blindly, except BaseException will work.
Both exceptions are no longer under Exception. This is to allow bare except clauses to act as a more viable default case by catching exceptions that inherit from Exception. With both KeyboardInterrupt and SystemExit acting as signals that the interpreter is expected to exit, catching them in the common case is the wrong semantics.
Inherits from Exception instead of from RuntimeError.
Originally inheriting from RuntimeError, NotImplementedError does not have any direct relation to the exception meant for use in user code as a quick-and-dirty exception. Thus it now directly inherits from Exception.
By requiring all objects passed to a raise statement to inherit from a specific superclass, all exceptions are guaranteed to have certain attributes. If PEP 344 is accepted, the attributes outlined there will be guaranteed to be on all exceptions raised. This should help facilitate debugging by making the querying of information from exceptions much easier.
The proposed hierarchy has BaseException as the required base class.
Enforcement is straightforward. Modifying RAISE_VARARGS to do an inheritance check first before raising an exception should be enough. For the C API, all functions that set an exception will have the same inheritance check applied.
In most existing Python 2.4 code, bare except clauses are too broad in the exceptions they catch. Typically only exceptions that signal an error are desired to be caught. This means that exceptions that are used to signify that the interpreter should exit should not be caught in the common case.
With KeyboardInterrupt and SystemExit moved to inherit from BaseException instead of Exception, changing bare except clauses to act as except Exception becomes a much more reasonable default. This change also will break very little code since these semantics are what most people want for bare except clauses.
The complete removal of bare except clauses has been argued for. The case has been made that they violate both Only One Way To Do It (OOWTDI) and Explicit Is Better Than Implicit (EIBTI) as listed in the Zen of Python. But Practicality Beats Purity (PBP), also in the Zen of Python, trumps both of these in this case. The BDFL has stated that bare except clauses will work this way .
The compiler will emit the bytecode for except Exception whenever a bare except clause is reached.
Because of the complexity and clutter that would be required to add all features planned in this PEP, the transition plan is very simple. In Python 2.5 BaseException is added. In Python 3.0, all remaining features (required superclass, change in inheritance, bare except clauses becoming the same as except Exception) will go into affect. In order to make all of this work in a backwards-compatible way in Python 2.5 would require very deep hacks in the exception machinery which could be error-prone and lead to a slowdown in performance for little benefit.
To help with the transition, the documentation will be changed to reflect several programming guidelines:
- When one wants to catch all exceptions, catch BaseException
- To catch all exceptions that do not represent the termination of the interpreter, catch Exception explicitly
- Explicitly catch KeyboardInterrupt and SystemExit; don't rely on inheritance from Exception to lead to the capture
- Always catch NotImplementedError explicitly instead of relying on the inheritance from RuntimeError
This was originally proposed because a DeprecationWarning can be viewed as a PendingDeprecationWarning that is being removed in the next version. But since enough people thought the inheritance could logically work the other way around, the idea was dropped.
Viewing attributes as part of the interface of a type caused the idea of inheriting from TypeError. But that partially defeats the thinking of duck typing and thus the idea was dropped.
Inheriting from NameError was suggested because objects can be viewed as having their own namespace where the attributes live and when an attribute is not found it is a namespace failure. This was also dropped as a possibility since not everyone shared this view.
Originally proposed based on the idea that EnvironmentError was an unneeded distinction, the BDFL overruled this idea .
Proposed to add symmetry to WindowsError, the BDFL said they won't be used enough . The idea of then removing WindowsError was proposed and accepted as reasonable, thus completely negating the idea of adding these exceptions.
Proposed because a SystemError is meant to lead to a system exit, the idea was removed since CriticalError indicates this better.
It has been suggested that ControlFlowException should inherit from Exception. This idea has been rejected based on the thinking that control flow exceptions typically do not all need to be caught by a single except clause.
NameError is considered more succinct and leaves open no possible mistyping of the capitalization of "Namespace" .
The thinking was that RuntimeError was in no way an obvious name for an exception meant to be used when a situation did not call for the creation of a new exception. The renaming was rejected on the basis that the exception is already used throughout the interpreter . Rejection of SimpleError was founded on the thought that people should be free to use whatever exception they choose and not have one so blatantly suggested .
Various renamings were suggested but non garnered more than a +0 vote (renaming ReferenceError to WeakReferenceError). The thinking was that the existing names were fine and no one had actively complained about them ever. To minimize backwards-compatibility issues and causing existing Python programmers extra pain, the renamings were removed.
The original thought was that since EOFError deals directly with I/O, it should subclass IOError. But since EOFError is used more as a signal that an event has occurred (the exhaustion of an I/O port), it should not subclass such a specific error exception.
Both classes deal with the interpreter, so why not have them have a common superclass? Because one of them means that the interpreter is in a state that it should not recover from while the other does not.
Grouping the deprecation warning exceptions together makes intuitive sense. But this sensical idea does not extend well when one considers how rarely either warning is used, let along at the same time.
Originally proposed based on the idea that having such a platform-specific exception should not be in the built-in namespace. It turns out, though, enough code exists that uses the exception to warrant it staying.
Proposed to make catching non-Exception inheriting exceptions easier along with easing the transition to the new hierarchy, the idea was rejected by the BDFL . The argument that existing code did not show enough instances of the pair of exceptions being caught and thus did not justify cluttering the built-in namespace was used.
Thanks to Robert Brewer, Josiah Carlson, Nick Coghlan, Timothy Delaney, Jack Diedrich, Fred L. Drake, Jr., Philip J. Eby, Greg Ewing, James Y. Knight, MA Lemburg, Guido van Rossum, Stephen J. Turnbull, Raymond Hettinger, and everyone else I missed for participating in the discussion.
|||python-dev Summary (An exception is an exception, unless it doesn't inherit from Exception) http://www.python.org/dev/summary/2004-08-01_2004-08-15.html#an-exception-is-an-exception-unless-it-doesn-t-inherit-from-exception|
|||python-dev email (PEP, take 2: Exception Reorganization for Python 3.0) https://mail.python.org/pipermail/python-dev/2005-August/055116.html|
|||exceptions module http://docs.python.org/library/exceptions.html|
|||python-dev thread (Pre-PEP: Exception Reorganization for Python 3.0) https://mail.python.org/pipermail/python-dev/2005-July/055020.html, https://mail.python.org/pipermail/python-dev/2005-August/055065.html|
|||python-dev thread (PEP, take 2: Exception Reorganization for Python 3.0) https://mail.python.org/pipermail/python-dev/2005-August/055103.html|
|||python-dev thread (Reorg PEP checked in) https://mail.python.org/pipermail/python-dev/2005-August/055138.html|
|||python-dev thread (Major revision of PEP 348 committed) https://mail.python.org/pipermail/python-dev/2005-August/055199.html|
|||python-dev thread (Exception Reorg PEP revised yet again) https://mail.python.org/pipermail/python-dev/2005-August/055292.html|
|||python-dev thread (PEP 348 (exception reorg) revised again) https://mail.python.org/pipermail/python-dev/2005-August/055412.html|
|||(1, 2) python-dev email (Pre-PEP: Exception Reorganization for Python 3.0) https://mail.python.org/pipermail/python-dev/2005-July/055019.html|
|||python-dev email (PEP, take 2: Exception Reorganization for Python 3.0) https://mail.python.org/pipermail/python-dev/2005-August/055159.html|
|||python-dev email (Exception Reorg PEP checked in) https://mail.python.org/pipermail/python-dev/2005-August/055149.html|
|||python-dev email (Exception Reorg PEP checked in) https://mail.python.org/pipermail/python-dev/2005-August/055175.html|
|||(1, 2) python-dev email (PEP 348 (exception reorg) revised again) https://mail.python.org/pipermail/python-dev/2005-August/055423.html|
|||Python Tutorial http://docs.python.org/tutorial/|
|||python-dev email (Bare except clauses in PEP 348) https://mail.python.org/pipermail/python-dev/2005-August/055676.html|
This document has been placed in the public domain.