PEP: 3136 Title: Labeled break and continue Version: $Revision$ Last-Modified: $Date$ Author: Matt Chisholm Status: Rejected Type: Standards Track Content-Type: text/x-rst Created: 30-Jun-2007 Python-Version: 3.1 Post-History: Rejection Notice ================ This PEP is rejected. See http://mail.python.org/pipermail/python-3000/2007-July/008663.html. Abstract ======== This PEP proposes support for labels in Python's ``break`` and ``continue`` statements. It is inspired by labeled ``break`` and ``continue`` in other languages, and the author's own infrequent but persistent need for such a feature. Introduction ============ The ``break`` statement allows the programmer to terminate a loop early, and the ``continue`` statement allows the programmer to move to the next iteration of a loop early. In Python currently, ``break`` and ``continue`` can apply only to the innermost enclosing loop. Adding support for labels to the ``break`` and ``continue`` statements is a logical extension to the existing behavior of the ``break`` and ``continue`` statements. Labeled ``break`` and ``continue`` can improve the readability and flexibility of complex code which uses nested loops. For brevity's sake, the examples and discussion in this PEP usually refers to the ``break`` statement. However, all of the examples and motivations apply equally to labeled ``continue``. Motivation ========== If the programmer wishes to move to the next iteration of an outer enclosing loop, or terminate multiple loops at once, he or she has a few less-than elegant options. Here's one common way of imitating labeled ``break`` in Python (For this and future examples, ``...`` denotes an arbitrary number of intervening lines of code):: for a in a_list: time_to_break_out_of_a = False ... for b in b_list: ... if condition_one(a, b): break ... if condition_two(a, b): time_to_break_out_of_a = True break ... if time_to_break_out_of_a: break ... This requires five lines and an extra variable, ``time_to_break_out_of_a``, to keep track of when to break out of the outer (a) loop. And those five lines are spread across many lines of code, making the control flow difficult to understand. This technique is also error-prone. A programmer modifying this code might inadvertently put new code after the end of the inner (b) loop but before the test for ``time_to_break_out_of_a``, instead of after the test. This means that code which should have been skipped by breaking out of the outer loop gets executed incorrectly. This could also be written with an exception. The programmer would declare a special exception, wrap the inner loop in a try, and catch the exception and break when you see it:: class BreakOutOfALoop(Exception): pass for a in a_list: ... try: for b in b_list: ... if condition_one(a, b): break ... if condition_two(a, b): raise BreakOutOfALoop ... except BreakOutOfALoop: break ... Again, though; this requires five lines and a new, single-purpose exception class (instead of a new variable), and spreads basic control flow out over many lines. And it breaks out of the inner loop with ``break`` and out of the other loop with an exception, which is inelegant. [#toowtdi]_ This next strategy might be the most elegant solution, assuming condition_two() is inexpensive to compute:: for a in a_list: ... for b in b_list: ... if condition_one(a, b): break ... if condition_two(a, b): break ... if condition_two(a, b) break ... Breaking twice is still inelegant. This implementation also relies on the fact that the inner (b) loop bleeds b into the outer for loop, which (although explicitly supported) is both surprising to novices, and in my opinion counter-intuitive and poor practice. The programmer must also still remember to put in both breaks on condition two and not insert code before the second break. A single conceptual action, breaking out of both loops on condition_two(), requires four lines of code at two indentation levels, possibly separated by many intervening lines at the end of the inner (b) loop. Other languages --------------- Now, put aside whatever dislike you may have for other programming languages, and consider the syntax of labeled ``break`` and ``continue``. In Perl:: ALOOP: foreach $a (@a_array){ ... BLOOP: foreach $b (@b_array){ ... if (condition_one($a,$b)){ last BLOOP; # same as plain old last; } ... if (condition_two($a,$b)){ last ALOOP; } ... } ... } (Notes: Perl uses ``last`` instead of ``break``. The BLOOP labels could be omitted; ``last`` and ``continue`` apply to the innermost loop by default.) PHP uses a number denoting the number of loops to break out of, rather than a label:: foreach ($a_array as $a){ .... foreach ($b_array as $b){ .... if (condition_one($a, $b)){ break 1; # same as plain old break } .... if (condition_two($a, $b)){ break 2; } .... } ... } C/C++, Java, and Ruby all have similar constructions. The control flow regarding when to break out of the outer (a) loop is fully encapsulated in the ``break`` statement which gets executed when the break condition is satisfied. The depth of the break statement does not matter. Control flow is not spread out. No extra variables, exceptions, or re-checking or storing of control conditions is required. There is no danger that code will get inadvertently inserted after the end of the inner (b) loop and before the break condition is re-checked inside the outer (a) loop. These are the benefits that labeled ``break`` and ``continue`` would bring to Python. What this PEP is not ==================== This PEP is not a proposal to add GOTO to Python. GOTO allows a programmer to jump to an arbitrary block or line of code, and generally makes control flow more difficult to follow. Although ``break`` and ``continue`` (with or without support for labels) can be considered a type of GOTO, it is much more restricted. Another Python construct, ``yield``, could also be considered a form of GOTO -- an even less restrictive one. The goal of this PEP is to propose an extension to the existing control flow tools ``break`` and ``continue``, to make control flow easier to understand, not more difficult. Labeled ``break`` and ``continue`` cannot transfer control to another function or method. They cannot even transfer control to an arbitrary line of code in the current scope. Currently, they can only affect the behavior of a loop, and are quite different and much more restricted than GOTO. This extension allows them to affect any enclosing loop in the current name-space, but it does not change their behavior to that of GOTO. Specification ============= Under all of these proposals, ``break`` and ``continue`` by themselves will continue to behave as they currently do, applying to the innermost loop by default. Proposal A - Explicit labels ---------------------------- The for and while loop syntax will be followed by an optional ``as`` or ``label`` (contextual) keyword [#keyword]_ and then an identifier, which may be used to identify the loop out of which to break (or which should be continued). The ``break`` (and ``continue``) statements will be followed by an optional identifier that refers to the loop out of which to break (or which should be continued). Here is an example using the ``as`` keyword:: for a in a_list as a_loop: ... for b in b_list as b_loop: ... if condition_one(a, b): break b_loop # same as plain old break ... if condition_two(a, b): break a_loop ... ... Or, with ``label`` instead of ``as``:: for a in a_list label a_loop: ... for b in b_list label b_loop: ... if condition_one(a, b): break b_loop # same as plain old break ... if condition_two(a, b): break a_loop ... ... This has all the benefits outlined above. It requires modifications to the language syntax: the syntax of ``break`` and ``continue`` syntax statements and for and while statements. It requires either a new conditional keyword ``label`` or an extension to the conditional keyword ``as``. [#as]_ It is unlikely to require any changes to existing Python programs. Passing an identifier not defined in the local scope to ``break`` or ``continue`` would raise a NameError. Proposal B - Numeric break & continue ------------------------------------- Rather than altering the syntax of ``for`` and ``while`` loops, ``break`` and ``continue`` would take a numeric argument denoting the enclosing loop which is being controlled, similar to PHP. It seems more Pythonic to me for ``break`` and ``continue`` to refer to loops indexing from zero, as opposed to indexing from one as PHP does. :: for a in a_list: ... for b in b_list: ... if condition_one(a,b): break 0 # same as plain old break ... if condition_two(a,b): break 1 ... ... Passing a number that was too large, or less than zero, or non-integer to ``break`` or ``continue`` would (probably) raise an IndexError. This proposal would not require any changes to existing Python programs. Proposal C - The reduplicative method ------------------------------------- The syntax of ``break`` and ``continue`` would be altered to allow multiple ``break`` and continue statements on the same line. Thus, ``break break`` would break out of the first and second enclosing loops. :: for a in a_list: ... for b in b_list: ... if condition_one(a,b): break # plain old break ... if condition_two(a,b): break break ... ... This would also allow the programmer to break out of the inner loop and continue the next outermost simply by writing ``break continue``, [#breakcontinue]_ and so on. I'm not sure what exception would be raised if the programmer used more ``break`` or ``continue`` statements than existing loops (perhaps a SyntaxError?). I expect this proposal to get rejected because it will be judged too difficult to understand. This proposal would not require any changes to existing Python programs. Proposal D - Explicit iterators ------------------------------- Rather than embellishing for and while loop syntax with labels, the programmer wishing to use labeled breaks would be required to create the iterator explicitly and assign it to a identifier if he or she wanted to ``break`` out of or ``continue`` that loop from within a deeper loop. :: a_iter = iter(a_list) for a in a_iter: ... b_iter = iter(b_list) for b in b_iter: ... if condition_one(a,b): break b_iter # same as plain old break ... if condition_two(a,b): break a_iter ... ... Passing a non-iterator object to ``break`` or ``continue`` would raise a TypeError; and a nonexistent identifier would raise a NameError. This proposal requires only one extra line to create a labeled loop, and no extra lines to break out of a containing loop, and no changes to existing Python programs. Proposal E - Explicit iterators and iterator methods ---------------------------------------------------- This is a variant of Proposal D. Iterators would need be created explicitly if anything other that the most basic use of ``break`` and ``continue`` was required. Instead of modifying the syntax of ``break`` and ``continue``, ``.break()`` and ``.continue()`` methods could be added to the Iterator type. :: a_iter = iter(a_list) for a in a_iter: ... b_iter = iter(b_list) for b in b_iter: ... if condition_one(a,b): b_iter.break() # same as plain old break ... if condition_two(a,b): a_iter.break() ... ... I expect that this proposal will get rejected on the grounds of sheer ugliness. However, it requires no changes to the language syntax whatsoever, nor does it require any changes to existing Python programs. Implementation ============== I have never looked at the Python language implementation itself, so I have no idea how difficult this would be to implement. If this PEP is accepted, but no one is available to write the feature, I will try to implement it myself. Footnotes ========= .. [#toowtdi] Breaking some loops with exceptions is inelegant because it's a violation of There's Only One Way To Do It. .. [#keyword] Or really any new contextual keyword that the community likes: ``as``, ``label``, ``labeled``, ``loop``, ``name``, ``named``, ``walrus``, whatever. .. [#as] The use of ``as`` in a similar context has been proposed here, http://sourceforge.net/tracker/index.php?func=detail&aid=1714448&group_id=5470&atid=355470 but to my knowledge this idea has not been written up as a PEP. .. [#breakcontinue] To continue the Nth outer loop, you would write break N-1 times and then continue. Only one ``continue`` would be allowed, and only at the end of a sequence of breaks. ``continue break`` or ``continue continue`` makes no sense. Resources ========= This issue has come up before, although it has never been resolved, to my knowledge. * `labeled breaks`__, on comp.lang.python, in the context of ``do...while`` loops __ http://groups.google.com/group/comp.lang.python/browse_thread/thread/6da848f762c9cf58/979ca3cd42633b52?lnk=gst&q=labeled+break&rnum=3#979ca3cd42633b52 * `break LABEL vs. exceptions + PROPOSAL`__, on python-list, as compared to using Exceptions for flow control __ http://mail.python.org/pipermail/python-list/1999-September/#11080 * `Named code blocks`__ on python-list, a suggestion motivated by the desire for labeled break / continue __ http://mail.python.org/pipermail/python-list/2001-April/#78439 * `mod_python bug fix`__ An example of someone setting a flag inside an inner loop that triggers a continue in the containing loop, to work around the absence of labeled break and continue __ http://mail-archives.apache.org/mod_mbox/httpd-python-cvs/200511.mbox/%3C20051112204322.4010.qmail@minotaur.apache.org%3E Copyright ========= This document has been placed in the public domain. .. Local Variables: mode: indented-text indent-tabs-mode: nil sentence-end-double-space: t fill-column: 70 coding: utf-8 End: